Up next in 10
In this episode of the Global AI October sessions we dive very deep into the world of advanced AI systems.
Show More Show Less View Video Transcript
0:07
[Music]
0:24
[Music]
0:31
[Music]
0:49
[Music]
1:22
go my door
1:24
go my door
1:24
go my door here we go hey
1:28
here we go hey
1:28
here we go hey alicia you're back hello
1:32
alicia you're back hello
1:32
alicia you're back hello we're we're at last episode
1:35
we're we're at last episode
1:35
we're we're at last episode i'm so so sad i mean
1:39
i'm so so sad i mean
1:40
i'm so so sad i mean about you october is my favoritest month
1:42
about you october is my favoritest month
1:42
about you october is my favoritest month and
1:43
and
1:43
and being part of october sessions has just
1:45
being part of october sessions has just
1:45
being part of october sessions has just been amazing for me
1:46
been amazing for me
1:46
been amazing for me so it's been a crazy month i have to say
1:51
so it's been a crazy month i have to say
1:51
so it's been a crazy month i have to say we had so many viewers and so many
1:53
we had so many viewers and so many
1:53
we had so many viewers and so many questions it's it's unbelievable
1:55
questions it's it's unbelievable
1:55
questions it's it's unbelievable actually
1:57
actually
1:57
actually and i'm happy i really enjoyed the
1:59
and i'm happy i really enjoyed the
1:59
and i'm happy i really enjoyed the session
2:01
session
2:01
session yeah i really did um i was out last week
2:05
yeah i really did um i was out last week
2:05
yeah i really did um i was out last week i was out sick
2:06
i was out sick
2:06
i was out sick but i tried to watch a few pieces of the
2:08
but i tried to watch a few pieces of the
2:08
but i tried to watch a few pieces of the stream last week and
2:09
stream last week and
2:09
stream last week and uh i was impressed actually by all the
2:11
uh i was impressed actually by all the
2:11
uh i was impressed actually by all the people that uh
2:13
people that uh
2:13
people that uh that actually jumped in for me to
2:15
that actually jumped in for me to
2:15
that actually jumped in for me to replace me but also about the guests
2:17
replace me but also about the guests
2:17
replace me but also about the guests that we had it was
2:18
that we had it was
2:18
that we had it was it was pretty cool yeah yeah
2:22
it was pretty cool yeah yeah
2:22
it was pretty cool yeah yeah so today we were actually talking about
2:25
so today we were actually talking about
2:25
so today we were actually talking about advanced ai
2:26
advanced ai
2:26
advanced ai but um thinking about this episode
2:29
but um thinking about this episode
2:29
but um thinking about this episode setting this up i actually have had sort
2:32
setting this up i actually have had sort
2:32
setting this up i actually have had sort of
2:32
of
2:32
of a hard time defining what is advanced
2:36
a hard time defining what is advanced
2:36
a hard time defining what is advanced and and i know you helped me so maybe
2:38
and and i know you helped me so maybe
2:38
and and i know you helped me so maybe you can explain what advanced actually
2:40
you can explain what advanced actually
2:40
you can explain what advanced actually means in this case
2:42
means in this case
2:42
means in this case well i i think that when we talk about
2:45
well i i think that when we talk about
2:45
well i i think that when we talk about ai
2:46
ai
2:46
ai um there's kind of two components to it
2:49
um there's kind of two components to it
2:49
um there's kind of two components to it one component is the tech
2:51
one component is the tech
2:51
one component is the tech and one component is to the business
2:54
and one component is to the business
2:54
and one component is to the business processes
2:55
processes
2:55
processes and i think that um there's a lot of use
2:58
and i think that um there's a lot of use
2:58
and i think that um there's a lot of use cases
2:59
cases
2:59
cases for machine learning and applying the
3:03
for machine learning and applying the
3:03
for machine learning and applying the mathematical principles to the data and
3:05
mathematical principles to the data and
3:05
mathematical principles to the data and the technology but when you apply the
3:07
the technology but when you apply the
3:08
the technology but when you apply the principles to enhance business processes
3:10
principles to enhance business processes
3:10
principles to enhance business processes and accelerate learning and accelerate
3:12
and accelerate learning and accelerate
3:12
and accelerate learning and accelerate turing
3:13
turing
3:13
turing of um notifications and alerting
3:16
of um notifications and alerting
3:16
of um notifications and alerting that's when your system is is said to be
3:19
that's when your system is is said to be
3:19
that's when your system is is said to be achieving
3:20
achieving
3:20
achieving that level of ai that level of
3:23
that level of ai that level of
3:23
that level of ai that level of independent
3:24
independent
3:24
independent knowledge and independent learning so um
3:28
knowledge and independent learning so um
3:28
knowledge and independent learning so um i'm very excited today because we are
3:30
i'm very excited today because we are
3:30
i'm very excited today because we are looking at
3:31
looking at
3:32
looking at a couple of systems where we're able to
3:35
a couple of systems where we're able to
3:35
a couple of systems where we're able to tie together different technologies so
3:37
tie together different technologies so
3:37
tie together different technologies so that we can help businesses
3:38
that we can help businesses
3:38
that we can help businesses accelerate their learning and accelerate
3:41
accelerate their learning and accelerate
3:41
accelerate their learning and accelerate their ability to respond and react
3:43
their ability to respond and react
3:43
their ability to respond and react to events yeah
3:46
to events yeah
3:46
to events yeah yeah to me that's pretty advanced i mean
3:50
yeah to me that's pretty advanced i mean
3:50
yeah to me that's pretty advanced i mean in our first episode we talked about
3:51
in our first episode we talked about
3:52
in our first episode we talked about using cognitive services that that's
3:53
using cognitive services that that's
3:53
using cognitive services that that's sort of the beginner level of ai to me
3:55
sort of the beginner level of ai to me
3:55
sort of the beginner level of ai to me it's very interesting to start with
3:58
it's very interesting to start with
3:58
it's very interesting to start with but it's also just the beginning of ai
4:01
but it's also just the beginning of ai
4:01
but it's also just the beginning of ai and
4:01
and
4:02
and once you get started building ai models
4:04
once you get started building ai models
4:04
once you get started building ai models then we we should talk about
4:05
then we we should talk about
4:05
then we we should talk about using it in a business process uh in in
4:08
using it in a business process uh in in
4:08
using it in a business process uh in in a business context actually
4:10
a business context actually
4:10
a business context actually um and then you run into a ton of
4:12
um and then you run into a ton of
4:12
um and then you run into a ton of problems that
4:13
problems that
4:14
problems that um that might force you to do something
4:17
um that might force you to do something
4:17
um that might force you to do something more advanced
4:18
more advanced
4:18
more advanced things like azure machine learning
4:19
things like azure machine learning
4:19
things like azure machine learning services and starting to use
4:21
services and starting to use
4:21
services and starting to use uh ml ops for that matter so that's
4:23
uh ml ops for that matter so that's
4:23
uh ml ops for that matter so that's actually part of our show today
4:25
actually part of our show today
4:25
actually part of our show today uh for those tuning in we're a little
4:26
uh for those tuning in we're a little
4:26
uh for those tuning in we're a little bit early that's on purpose we're
4:28
bit early that's on purpose we're
4:28
bit early that's on purpose we're sound checking we're making sure that we
4:30
sound checking we're making sure that we
4:30
sound checking we're making sure that we have a lot of fun together and talk
4:32
have a lot of fun together and talk
4:32
have a lot of fun together and talk about ai
4:33
about ai
4:33
about ai so you're part of our back talk so to
4:35
so you're part of our back talk so to
4:35
so you're part of our back talk so to speak
4:37
speak
4:37
speak so yeah advanced ai we've got some
4:40
so yeah advanced ai we've got some
4:40
so yeah advanced ai we've got some interesting stuff coming up around
4:45
augmented reality generating office
4:47
augmented reality generating office
4:47
augmented reality generating office floors we talked about this before
4:49
floors we talked about this before
4:49
floors we talked about this before a few episodes ago and i'm happy to to
4:52
a few episodes ago and i'm happy to to
4:52
a few episodes ago and i'm happy to to say that we have people who are working
4:54
say that we have people who are working
4:54
say that we have people who are working on this stuff
4:55
on this stuff
4:55
on this stuff and you found those people this thank
4:58
and you found those people this thank
4:58
and you found those people this thank you well you know
4:59
you well you know
4:59
you well you know one it's wonderful working with you
5:01
one it's wonderful working with you
5:01
one it's wonderful working with you because you know when we had our session
5:04
because you know when we had our session
5:04
because you know when we had our session on our custom vision session two weeks
5:06
on our custom vision session two weeks
5:06
on our custom vision session two weeks ago and we were talking about
5:08
ago and we were talking about
5:08
ago and we were talking about you know well what what is it going to
5:10
you know well what what is it going to
5:10
you know well what what is it going to take to make alicia happy
5:12
take to make alicia happy
5:12
take to make alicia happy and you really took that to heart and
5:15
and you really took that to heart and
5:15
and you really took that to heart and you really
5:15
you really
5:15
you really challenged yourself to figure out you
5:18
challenged yourself to figure out you
5:18
challenged yourself to figure out you know
5:19
know
5:19
know how to find alicia more windows space
5:22
how to find alicia more windows space
5:22
how to find alicia more windows space and more trees to make her happy and you
5:25
and more trees to make her happy and you
5:25
and more trees to make her happy and you know
5:26
know
5:26
know part of solving that problem is
5:27
part of solving that problem is
5:27
part of solving that problem is absolutely locating the technology
5:29
absolutely locating the technology
5:30
absolutely locating the technology so i think it's great that you know you
5:32
so i think it's great that you know you
5:32
so i think it's great that you know you were supportive
5:33
were supportive
5:33
were supportive and uh seeking out a solution to that
5:36
and uh seeking out a solution to that
5:36
and uh seeking out a solution to that problem and um
5:38
problem and um
5:38
problem and um so i i'm really excited to to have
5:42
so i i'm really excited to to have
5:42
so i i'm really excited to to have um fergus kidd and jay
5:45
um fergus kidd and jay
5:45
um fergus kidd and jay nadarajan from from avanade here today
5:49
nadarajan from from avanade here today
5:49
nadarajan from from avanade here today to to talk to us about some of these
5:51
to to talk to us about some of these
5:51
to to talk to us about some of these solutions
5:52
solutions
5:52
solutions yeah and the other advanced topic that
5:55
yeah and the other advanced topic that
5:55
yeah and the other advanced topic that we also have this episode is uh noelle
5:57
we also have this episode is uh noelle
5:57
we also have this episode is uh noelle silver
5:58
silver
5:58
silver she is going to talk about
5:59
she is going to talk about
5:59
she is going to talk about explainability and models
6:01
explainability and models
6:01
explainability and models i'm very curious about that it's all
6:02
i'm very curious about that it's all
6:02
i'm very curious about that it's all responsible ai and people may have
6:04
responsible ai and people may have
6:04
responsible ai and people may have noticed in the last
6:06
noticed in the last
6:06
noticed in the last few episodes that we're very fond of
6:08
few episodes that we're very fond of
6:08
few episodes that we're very fond of that topic
6:09
that topic
6:09
that topic i think it's important because it helps
6:11
i think it's important because it helps
6:11
i think it's important because it helps you prevent
6:12
you prevent
6:12
you prevent certain disasters that that you may run
6:16
certain disasters that that you may run
6:16
certain disasters that that you may run into in ai
6:18
into in ai
6:18
into in ai and today we're going to talk about how
6:20
and today we're going to talk about how
6:20
and today we're going to talk about how do you actually explain this
6:21
do you actually explain this
6:21
do you actually explain this black box of ai it's pretty advanced
6:23
black box of ai it's pretty advanced
6:23
black box of ai it's pretty advanced stuff uh but it's very very useful to to
6:26
stuff uh but it's very very useful to to
6:26
stuff uh but it's very very useful to to use that in your application
6:28
use that in your application
6:28
use that in your application and um i should also note so we have
6:30
and um i should also note so we have
6:30
and um i should also note so we have seth soraris he's been here before
6:32
seth soraris he's been here before
6:32
seth soraris he's been here before in episode one and today he's going to
6:36
in episode one and today he's going to
6:36
in episode one and today he's going to show us
6:36
show us
6:36
show us uh deep learning with pytorch so
6:38
uh deep learning with pytorch so
6:38
uh deep learning with pytorch so building neural networks with pytorch
6:41
building neural networks with pytorch
6:41
building neural networks with pytorch and i can hear people groan
6:44
and i can hear people groan
6:44
and i can hear people groan that's not very advanced well i asked
6:47
that's not very advanced well i asked
6:47
that's not very advanced well i asked him specifically to give us an advanced
6:49
him specifically to give us an advanced
6:49
him specifically to give us an advanced session and he asked me
6:51
session and he asked me
6:51
session and he asked me is it phd advanced or program um
6:55
is it phd advanced or program um
6:55
is it phd advanced or program um uh yeah well we go we're going for
6:57
uh yeah well we go we're going for
6:57
uh yeah well we go we're going for programmer advanced but if people have
6:59
programmer advanced but if people have
6:59
programmer advanced but if people have questions for him
7:00
questions for him
7:00
questions for him that are a phd advance he's happy to
7:02
that are a phd advance he's happy to
7:02
that are a phd advance he's happy to answer those two
7:04
answer those two
7:04
answer those two so this should be exciting
7:08
exciting episode and we're going out
7:12
exciting episode and we're going out
7:12
exciting episode and we're going out with a bang so to say
7:18
so i'm so excited for today's session
7:20
so i'm so excited for today's session
7:20
so i'm so excited for today's session yeah
7:21
yeah
7:21
yeah shall we get started for real i mean we
7:24
shall we get started for real i mean we
7:24
shall we get started for real i mean we it's seven o'clock so it's time time to
7:27
it's seven o'clock so it's time time to
7:27
it's seven o'clock so it's time time to to run our first guest um
7:30
to run our first guest um
7:30
to run our first guest um people are holding up piece of pieces of
7:32
people are holding up piece of pieces of
7:32
people are holding up piece of pieces of paper
7:33
paper
7:33
paper interpretation
7:37
so alicia you mentioned that you have
7:41
so alicia you mentioned that you have
7:41
so alicia you mentioned that you have fergus on the show and jay
7:42
fergus on the show and jay
7:42
fergus on the show and jay can you maybe uh introduce them to us
7:45
can you maybe uh introduce them to us
7:45
can you maybe uh introduce them to us they're colleagues of yours so
7:47
they're colleagues of yours so
7:47
they're colleagues of yours so uh yeah let's see what they've got
7:51
uh yeah let's see what they've got
7:51
uh yeah let's see what they've got hi so good afternoon and welcome to jay
7:55
hi so good afternoon and welcome to jay
7:55
hi so good afternoon and welcome to jay and to fergus so jay is one of our
7:57
and to fergus so jay is one of our
7:57
and to fergus so jay is one of our executive leaders in data in
7:59
executive leaders in data in
7:59
executive leaders in data in ai um she did spend um
8:02
ai um she did spend um
8:02
ai um she did spend um a lot of time at microsoft um leaning
8:05
a lot of time at microsoft um leaning
8:05
a lot of time at microsoft um leaning out
8:06
out
8:06
out their i believe is data in aeon
8:10
their i believe is data in aeon
8:10
their i believe is data in aeon and uh fergus is part of avanade's coe
8:14
and uh fergus is part of avanade's coe
8:14
and uh fergus is part of avanade's coe so that's our center of excellence
8:17
so that's our center of excellence
8:17
so that's our center of excellence program
8:17
program
8:17
program and he gets to work on all the fun stuff
8:21
and he gets to work on all the fun stuff
8:21
and he gets to work on all the fun stuff like when i'm reincarnated i want to be
8:23
like when i'm reincarnated i want to be
8:23
like when i'm reincarnated i want to be reincarnated as fergus because
8:26
reincarnated as fergus because
8:26
reincarnated as fergus because they give him like you want to talk
8:28
they give him like you want to talk
8:28
they give him like you want to talk about the kid with the coolest
8:30
about the kid with the coolest
8:30
about the kid with the coolest tools or the coolest toys that is fergus
8:33
tools or the coolest toys that is fergus
8:33
tools or the coolest toys that is fergus they let him play with the latest and
8:35
they let him play with the latest and
8:35
they let him play with the latest and greatest and
8:36
greatest and
8:36
greatest and um i'm so grateful that he's accepted
8:40
um i'm so grateful that he's accepted
8:40
um i'm so grateful that he's accepted our
8:40
our
8:40
our our offer to speak today and i'm happy
8:43
our offer to speak today and i'm happy
8:43
our offer to speak today and i'm happy to welcome him
8:43
to welcome him
8:44
to welcome him into the community and i i look forward
8:45
into the community and i i look forward
8:45
into the community and i i look forward to hearing more from
8:47
to hearing more from
8:47
to hearing more from from both of you so welcome
8:50
from both of you so welcome
8:50
from both of you so welcome cool thanks great to be here thank you
8:52
cool thanks great to be here thank you
8:52
cool thanks great to be here thank you so if people have questions
8:54
so if people have questions
8:54
so if people have questions and i bet they will have about this
8:56
and i bet they will have about this
8:56
and i bet they will have about this topic feel free to tweet to us
8:58
topic feel free to tweet to us
8:58
topic feel free to tweet to us um ask them in the live chat we have a
9:00
um ask them in the live chat we have a
9:00
um ask them in the live chat we have a live chat and chat on our website
9:02
live chat and chat on our website
9:02
live chat and chat on our website um and we'll we'll get them to the
9:04
um and we'll we'll get them to the
9:04
um and we'll we'll get them to the speakers tonight
9:06
speakers tonight
9:06
speakers tonight i should also note if if you're
9:08
i should also note if if you're
9:08
i should also note if if you're interested in winning a
9:10
interested in winning a
9:10
interested in winning a 50 gift card we're giving one away the
9:12
50 gift card we're giving one away the
9:12
50 gift card we're giving one away the last one of the series
9:13
last one of the series
9:14
last one of the series um so you can win one if you tweet to us
9:17
um so you can win one if you tweet to us
9:17
um so you can win one if you tweet to us using the hashtag c sharp corner and the
9:18
using the hashtag c sharp corner and the
9:18
using the hashtag c sharp corner and the hashtag global ai community
9:21
hashtag global ai community
9:21
hashtag global ai community and uh be sure to fill out the form that
9:22
and uh be sure to fill out the form that
9:22
and uh be sure to fill out the form that we're posting in the live chat so we
9:24
we're posting in the live chat so we
9:24
we're posting in the live chat so we know you've tweeted and we don't have to
9:26
know you've tweeted and we don't have to
9:26
know you've tweeted and we don't have to search all twitter for
9:27
search all twitter for
9:27
search all twitter for uh tweets um and you can make a chance
9:30
uh tweets um and you can make a chance
9:30
uh tweets um and you can make a chance and we'll announce the winner at the end
9:31
and we'll announce the winner at the end
9:31
and we'll announce the winner at the end of the show so that's that's good to
9:33
of the show so that's that's good to
9:33
of the show so that's that's good to know
9:33
know
9:33
know um yeah take it away fergus
9:38
awesome so i i'll start off by uh
9:40
awesome so i i'll start off by uh
9:40
awesome so i i'll start off by uh introducing uh
9:41
introducing uh
9:41
introducing uh jay who's gonna talk us through the
9:43
jay who's gonna talk us through the
9:43
jay who's gonna talk us through the first couple of slides so
9:45
first couple of slides so
9:45
first couple of slides so um over to you jay all right thank you
9:48
um over to you jay all right thank you
9:48
um over to you jay all right thank you uh
9:48
uh
9:48
uh it's a pleasure to be here and thank you
9:51
it's a pleasure to be here and thank you
9:51
it's a pleasure to be here and thank you all for joining us today
9:53
all for joining us today
9:53
all for joining us today um as alicia introduce me i'm jayna
9:56
um as alicia introduce me i'm jayna
9:56
um as alicia introduce me i'm jayna tarajin
9:57
tarajin
9:57
tarajin and i lead the data on ai for west
9:59
and i lead the data on ai for west
9:59
and i lead the data on ai for west region for avanade and prior to avenatta
10:02
region for avanade and prior to avenatta
10:02
region for avanade and prior to avenatta i
10:02
i
10:02
i spent about 12 years in microsoft so
10:05
spent about 12 years in microsoft so
10:06
spent about 12 years in microsoft so uh it is awesome to be here and
10:10
uh it is awesome to be here and
10:10
uh it is awesome to be here and um as we are thinking about data and the
10:12
um as we are thinking about data and the
10:12
um as we are thinking about data and the ai
10:13
ai
10:13
ai i i i as you all know
10:18
i i i as you all know
10:18
i i i as you all know with the pandemic and with the
10:21
with the pandemic and with the
10:21
with the pandemic and with the technology shift that we are seeing
10:23
technology shift that we are seeing
10:23
technology shift that we are seeing we are slowly coming to realization that
10:26
we are slowly coming to realization that
10:26
we are slowly coming to realization that the technology
10:27
the technology
10:27
the technology shift that we are seeing is going to be
10:29
shift that we are seeing is going to be
10:29
shift that we are seeing is going to be pertinent
10:30
pertinent
10:30
pertinent in a lot of customer conversations when
10:33
in a lot of customer conversations when
10:33
in a lot of customer conversations when i go
10:33
i go
10:34
i go in most companies and we start talking
10:36
in most companies and we start talking
10:36
in most companies and we start talking about some of these technologies and
10:38
about some of these technologies and
10:38
about some of these technologies and how can you derive insight using data
10:42
how can you derive insight using data
10:42
how can you derive insight using data and how can you actually go from
10:45
and how can you actually go from
10:45
and how can you actually go from descriptive to predictive to
10:47
descriptive to predictive to
10:47
descriptive to predictive to prescriptive
10:48
prescriptive
10:48
prescriptive analytics most companies are asking is
10:51
analytics most companies are asking is
10:51
analytics most companies are asking is this technology
10:52
this technology
10:52
this technology new and you know how do we explore the
10:54
new and you know how do we explore the
10:54
new and you know how do we explore the art of the possible
10:56
art of the possible
10:56
art of the possible so that is not on one side and if
11:00
so that is not on one side and if
11:00
so that is not on one side and if we look at the organizations
11:03
we look at the organizations
11:03
we look at the organizations in general they're
11:06
in general they're
11:06
in general they're they're slowly recognizing the white
11:08
they're slowly recognizing the white
11:08
they're slowly recognizing the white spaces and the blind spots
11:10
spaces and the blind spots
11:10
spaces and the blind spots and they want to do something about
11:13
and they want to do something about
11:13
and they want to do something about how they can operate in this new world
11:16
how they can operate in this new world
11:16
how they can operate in this new world um all that all the talents are
11:19
um all that all the talents are
11:19
um all that all the talents are right now disconnected and and you can
11:22
right now disconnected and and you can
11:22
right now disconnected and and you can you can actually
11:24
you can actually
11:24
you can actually live anywhere and you can work from
11:26
live anywhere and you can work from
11:26
live anywhere and you can work from anywhere as well so as long
11:28
anywhere as well so as long
11:28
anywhere as well so as long as you are technically savvy the
11:30
as you are technically savvy the
11:30
as you are technically savvy the opportunities are actually
11:32
opportunities are actually
11:32
opportunities are actually endless even with
11:35
endless even with
11:35
endless even with all the talents that are available in
11:37
all the talents that are available in
11:37
all the talents that are available in all the resources that are available
11:39
all the resources that are available
11:39
all the resources that are available think about the advancements that we
11:41
think about the advancements that we
11:41
think about the advancements that we have in
11:41
have in
11:41
have in hardware algorithms and data and
11:43
hardware algorithms and data and
11:44
hardware algorithms and data and everything that came with the fourth
11:45
everything that came with the fourth
11:45
everything that came with the fourth industrial revolution
11:47
industrial revolution
11:47
industrial revolution um a lot of organizations that
11:50
um a lot of organizations that
11:50
um a lot of organizations that we work with are do not still use the
11:54
we work with are do not still use the
11:54
we work with are do not still use the data
11:54
data
11:54
data to actually make the decisions this was
11:57
to actually make the decisions this was
11:57
to actually make the decisions this was one of the statistics that i was
11:59
one of the statistics that i was
11:59
one of the statistics that i was reading a while ago and i i forget
12:02
reading a while ago and i i forget
12:02
reading a while ago and i i forget where i actually read this but the
12:05
where i actually read this but the
12:05
where i actually read this but the statistics go somewhere
12:07
statistics go somewhere
12:07
statistics go somewhere something like this actually one in
12:09
something like this actually one in
12:09
something like this actually one in three business leaders
12:10
three business leaders
12:10
three business leaders make critical decisions without the
12:13
make critical decisions without the
12:13
make critical decisions without the information they actually need
12:16
information they actually need
12:16
information they actually need and even when the decision makers are
12:18
and even when the decision makers are
12:18
and even when the decision makers are making decisions they first lean into a
12:20
making decisions they first lean into a
12:20
making decisions they first lean into a personal experience
12:22
personal experience
12:22
personal experience then they use analytics and lastly
12:25
then they use analytics and lastly
12:25
then they use analytics and lastly they go to a collective experience
12:27
they go to a collective experience
12:27
they go to a collective experience before they make a decision
12:30
before they make a decision
12:30
before they make a decision and you may actually be wondering why am
12:33
and you may actually be wondering why am
12:33
and you may actually be wondering why am i talking about this
12:34
i talking about this
12:34
i talking about this this becomes very important because you
12:36
this becomes very important because you
12:36
this becomes very important because you need to have the data
12:38
need to have the data
12:38
need to have the data for you to make the right decisions for
12:40
for you to make the right decisions for
12:40
for you to make the right decisions for you to predict
12:42
you to predict
12:42
you to predict how can you actually use technology to
12:45
how can you actually use technology to
12:45
how can you actually use technology to make the decisions that you want to make
12:48
make the decisions that you want to make
12:48
make the decisions that you want to make so actually gathering the data
12:50
so actually gathering the data
12:50
so actually gathering the data collecting it and storing it
12:52
collecting it and storing it
12:52
collecting it and storing it in a cloud or storing it
12:56
in a cloud or storing it
12:56
in a cloud or storing it in in in a location where you have an
12:59
in in in a location where you have an
12:59
in in in a location where you have an ability to scale
13:00
ability to scale
13:00
ability to scale is going to become super important
13:03
is going to become super important
13:03
is going to become super important for organizations to leapfrog into this
13:06
for organizations to leapfrog into this
13:06
for organizations to leapfrog into this new world
13:09
on the same topic as
13:13
on the same topic as
13:13
on the same topic as as you all know when when the code
13:15
as you all know when when the code
13:15
as you all know when when the code actually started
13:18
actually started
13:18
actually started we all knew that you know we have to
13:21
we all knew that you know we have to
13:21
we all knew that you know we have to have the right data and we
13:22
have the right data and we
13:22
have the right data and we when the ai would help us to analyze the
13:24
when the ai would help us to analyze the
13:24
when the ai would help us to analyze the data better so that we can get to the
13:26
data better so that we can get to the
13:26
data better so that we can get to the right
13:27
right
13:27
right outcome however
13:30
outcome however
13:30
outcome however kovic taught us a very important lesson
13:32
kovic taught us a very important lesson
13:32
kovic taught us a very important lesson on the importance of data and
13:34
on the importance of data and
13:34
on the importance of data and why interpreting the data across the
13:37
why interpreting the data across the
13:37
why interpreting the data across the whole value chain becomes
13:39
whole value chain becomes
13:39
whole value chain becomes important because when initially when
13:42
important because when initially when
13:42
important because when initially when the reports and everything started to
13:44
the reports and everything started to
13:44
the reports and everything started to came out
13:45
came out
13:45
came out nobody was on the same page none of the
13:47
nobody was on the same page none of the
13:47
nobody was on the same page none of the analysis actually matched
13:50
analysis actually matched
13:50
analysis actually matched and that is because the the data was
13:53
and that is because the the data was
13:54
and that is because the the data was inconsistent
13:55
inconsistent
13:55
inconsistent and not all the data was actually
13:57
and not all the data was actually
13:57
and not all the data was actually available
13:58
available
13:58
available and most of the cities were actually
14:01
and most of the cities were actually
14:01
and most of the cities were actually dealing with the digitizing the data
14:03
dealing with the digitizing the data
14:03
dealing with the digitizing the data and once the data was digitized it
14:06
and once the data was digitized it
14:06
and once the data was digitized it needed to be cleaned and governed and
14:08
needed to be cleaned and governed and
14:08
needed to be cleaned and governed and everything before you can
14:09
everything before you can
14:10
everything before you can actually have any kind of insights
14:12
actually have any kind of insights
14:12
actually have any kind of insights derived from that data
14:15
derived from that data
14:15
derived from that data on the flip side of it we
14:18
on the flip side of it we
14:18
on the flip side of it we had a real need for how can we
14:21
had a real need for how can we
14:21
had a real need for how can we use this technology to make people safer
14:25
use this technology to make people safer
14:25
use this technology to make people safer for example when the cdc started
14:27
for example when the cdc started
14:27
for example when the cdc started mandating that
14:28
mandating that
14:28
mandating that you know hey it's they didn't mandate
14:30
you know hey it's they didn't mandate
14:30
you know hey it's they didn't mandate they were recommending that you need to
14:32
they were recommending that you need to
14:32
they were recommending that you need to be six feet apart
14:34
be six feet apart
14:34
be six feet apart when you were in a public space
14:37
when you were in a public space
14:38
when you were in a public space it it became important that how do we
14:40
it it became important that how do we
14:40
it it became important that how do we actually measure
14:42
actually measure
14:42
actually measure we can use the cameras we can use it we
14:45
we can use the cameras we can use it we
14:45
we can use the cameras we can use it we can use the technology that was
14:46
can use the technology that was
14:46
can use the technology that was available
14:47
available
14:47
available to make it smart and understand are
14:51
to make it smart and understand are
14:51
to make it smart and understand are people following the guidelines that
14:53
people following the guidelines that
14:54
people following the guidelines that that cdc is actually placing so that
14:57
that cdc is actually placing so that
14:57
that cdc is actually placing so that we can get the workers back to the work
14:59
we can get the workers back to the work
15:00
we can get the workers back to the work safely
15:00
safely
15:00
safely and securely as well so those are some
15:04
and securely as well so those are some
15:04
and securely as well so those are some of the areas where we started mixing the
15:06
of the areas where we started mixing the
15:06
of the areas where we started mixing the technology we can actually look at the
15:08
technology we can actually look at the
15:08
technology we can actually look at the cameras and we can see the data that we
15:10
cameras and we can see the data that we
15:10
cameras and we can see the data that we get from the cameras and see you know
15:12
get from the cameras and see you know
15:12
get from the cameras and see you know are people actually six feet apart or
15:14
are people actually six feet apart or
15:14
are people actually six feet apart or not if they're not
15:16
not if they're not
15:16
not if they're not how can we put some guidance around it
15:18
how can we put some guidance around it
15:18
how can we put some guidance around it to make sure that people
15:20
to make sure that people
15:20
to make sure that people are standing safely so this is a very
15:22
are standing safely so this is a very
15:22
are standing safely so this is a very very
15:23
very
15:23
very simple use case that i'm talking about
15:25
simple use case that i'm talking about
15:25
simple use case that i'm talking about how can
15:26
how can
15:26
how can we actually make or how can we leverage
15:29
we actually make or how can we leverage
15:29
we actually make or how can we leverage the smart spaces
15:32
the smart spaces
15:32
the smart spaces and how can we really have a huge
15:35
and how can we really have a huge
15:35
and how can we really have a huge impact on our life because whenever we
15:39
impact on our life because whenever we
15:39
impact on our life because whenever we talk about ai
15:40
talk about ai
15:40
talk about ai or mixed reality extended reality
15:43
or mixed reality extended reality
15:43
or mixed reality extended reality however you want to phrase it as
15:45
however you want to phrase it as
15:45
however you want to phrase it as some of them are very visible if you
15:48
some of them are very visible if you
15:48
some of them are very visible if you look at the autonomous vehicle it's very
15:50
look at the autonomous vehicle it's very
15:50
look at the autonomous vehicle it's very very visible on how technology actually
15:53
very visible on how technology actually
15:53
very visible on how technology actually helps you but a lot of these smart
15:55
helps you but a lot of these smart
15:55
helps you but a lot of these smart spaces the technology that are used in
15:57
spaces the technology that are used in
15:57
spaces the technology that are used in the smart spaces
15:58
the smart spaces
15:58
the smart spaces are very less apparent but they're very
16:01
are very less apparent but they're very
16:01
are very less apparent but they're very capable they're very efficient and
16:03
capable they're very efficient and
16:03
capable they're very efficient and they're very necessary services that
16:05
they're very necessary services that
16:05
they're very necessary services that is going to help us transform our lives
16:08
is going to help us transform our lives
16:08
is going to help us transform our lives as well
16:09
as well
16:09
as well so with that uh we can go to the next
16:13
so with that uh we can go to the next
16:13
so with that uh we can go to the next slide
16:14
slide
16:14
slide so i i just want you to leave with the
16:17
so i i just want you to leave with the
16:17
so i i just want you to leave with the one thought before i transition to
16:19
one thought before i transition to
16:19
one thought before i transition to fergus
16:21
fergus
16:21
fergus ai is going to impact your business
16:24
ai is going to impact your business
16:24
ai is going to impact your business and that is very very certain and
16:27
and that is very very certain and
16:27
and that is very very certain and it's actually up to us to decide how are
16:30
it's actually up to us to decide how are
16:30
it's actually up to us to decide how are we going to adapt
16:32
we going to adapt
16:32
we going to adapt meaning how are we going to scale up you
16:35
meaning how are we going to scale up you
16:35
meaning how are we going to scale up you can look at healthcare as one of the
16:36
can look at healthcare as one of the
16:36
can look at healthcare as one of the industries where
16:38
industries where
16:38
industries where ai is actually transforming people's
16:41
ai is actually transforming people's
16:41
ai is actually transforming people's lives
16:42
lives
16:42
lives that's one area manufacturing and even
16:45
that's one area manufacturing and even
16:46
that's one area manufacturing and even construction
16:47
construction
16:47
construction is becoming one of the one of the
16:49
is becoming one of the one of the
16:49
is becoming one of the one of the industries where ai
16:50
industries where ai
16:50
industries where ai is being adopted much faster so
16:54
is being adopted much faster so
16:54
is being adopted much faster so so for us to be for us to be
16:58
so for us to be for us to be
16:58
so for us to be for us to be in the business or for us to transform
17:00
in the business or for us to transform
17:00
in the business or for us to transform our
17:01
our
17:01
our reality we need to actually redefine
17:05
reality we need to actually redefine
17:05
reality we need to actually redefine how ai is going to help us do things
17:07
how ai is going to help us do things
17:07
how ai is going to help us do things differently
17:08
differently
17:08
differently and how it is how how it is going to
17:11
and how it is how how it is going to
17:11
and how it is how how it is going to help us do different things as well
17:13
help us do different things as well
17:13
help us do different things as well meaning how can we leave some of the
17:17
meaning how can we leave some of the
17:17
meaning how can we leave some of the some of the mundane tasks to ai so that
17:20
some of the mundane tasks to ai so that
17:20
some of the mundane tasks to ai so that we can move to a much productive
17:23
we can move to a much productive
17:23
we can move to a much productive challenges so with that i'm going to
17:26
challenges so with that i'm going to
17:26
challenges so with that i'm going to leave it to fergus transition to fergus
17:28
leave it to fergus transition to fergus
17:28
leave it to fergus transition to fergus so that he can start
17:29
so that he can start
17:30
so that he can start um showcasing some of the awesome work
17:32
um showcasing some of the awesome work
17:32
um showcasing some of the awesome work that we did around smart spaces
17:34
that we did around smart spaces
17:34
that we did around smart spaces fergus great thanks so much jay
17:37
fergus great thanks so much jay
17:37
fergus great thanks so much jay um and thanks alicia for the uh epic
17:40
um and thanks alicia for the uh epic
17:40
um and thanks alicia for the uh epic introduction
17:41
introduction
17:41
introduction so my name is uh fergus kidd
17:44
so my name is uh fergus kidd
17:44
so my name is uh fergus kidd i am one of the emerging technology
17:47
i am one of the emerging technology
17:47
i am one of the emerging technology engineers at avanade based in
17:49
engineers at avanade based in
17:49
engineers at avanade based in london so the emerging technology team
17:53
london so the emerging technology team
17:53
london so the emerging technology team at avanide
17:54
at avanide
17:54
at avanide at avanade looks at a really broad range
17:56
at avanade looks at a really broad range
17:56
at avanade looks at a really broad range of technologies that we think are sort
17:58
of technologies that we think are sort
17:58
of technologies that we think are sort of coming up and uh of
18:00
of coming up and uh of
18:00
of coming up and uh of interest and importance to our clients
18:02
interest and importance to our clients
18:02
interest and importance to our clients in the near future so we look at
18:03
in the near future so we look at
18:04
in the near future so we look at everything from quantum computing as you
18:05
everything from quantum computing as you
18:05
everything from quantum computing as you see on the left
18:06
see on the left
18:06
see on the left physical robots and between our physical
18:08
physical robots and between our physical
18:08
physical robots and between our physical spaces research
18:09
spaces research
18:10
spaces research 5g next generation ai all sorts of
18:13
5g next generation ai all sorts of
18:13
5g next generation ai all sorts of different topics
18:13
different topics
18:14
different topics uh so i do have a one of the cooler jobs
18:16
uh so i do have a one of the cooler jobs
18:16
uh so i do have a one of the cooler jobs here in avenard
18:17
here in avenard
18:17
here in avenard um i would be the first person to admit
18:19
um i would be the first person to admit
18:19
um i would be the first person to admit that for sure
18:21
that for sure
18:21
that for sure so what i want to talk to you today is
18:25
so what i want to talk to you today is
18:25
so what i want to talk to you today is i want to introduce a couple of
18:26
i want to introduce a couple of
18:26
i want to introduce a couple of technical topics just in case you're not
18:28
technical topics just in case you're not
18:28
technical topics just in case you're not familiar with those and just define them
18:30
familiar with those and just define them
18:30
familiar with those and just define them in the way that
18:31
in the way that
18:31
in the way that we do here at avanade um i want to
18:34
we do here at avanade um i want to
18:34
we do here at avanade um i want to talk briefly through what we're actually
18:36
talk briefly through what we're actually
18:36
talk briefly through what we're actually doing with those technology
18:38
doing with those technology
18:38
doing with those technology and then give you a demo of what we've
18:40
and then give you a demo of what we've
18:40
and then give you a demo of what we've actually done so far what that looks
18:41
actually done so far what that looks
18:41
actually done so far what that looks like
18:41
like
18:42
like and what the kind of results are around
18:44
and what the kind of results are around
18:44
and what the kind of results are around this idea of like physical spaces
18:46
this idea of like physical spaces
18:46
this idea of like physical spaces um and and
18:49
um and and
18:49
um and and um extended reality so the first thing
18:53
um extended reality so the first thing
18:53
um extended reality so the first thing it will make a little bit more sense as
18:54
it will make a little bit more sense as
18:54
it will make a little bit more sense as we go through why i'm talking about
18:55
we go through why i'm talking about
18:55
we go through why i'm talking about digital twins
18:57
digital twins
18:57
digital twins but firstly i'm just going to go through
18:58
but firstly i'm just going to go through
18:58
but firstly i'm just going to go through what a digital twin is so digital twin
19:00
what a digital twin is so digital twin
19:00
what a digital twin is so digital twin is a virtual replica
19:02
is a virtual replica
19:02
is a virtual replica of a physical entity it could be
19:04
of a physical entity it could be
19:04
of a physical entity it could be anything it could be your car it could
19:05
anything it could be your car it could
19:05
anything it could be your car it could be your mobile phone
19:07
be your mobile phone
19:07
be your mobile phone anything uh physical small large complex
19:12
anything uh physical small large complex
19:12
anything uh physical small large complex and included in the digital twin is
19:14
and included in the digital twin is
19:14
and included in the digital twin is first off a data model
19:16
first off a data model
19:16
first off a data model so today we're going to talk about smart
19:18
so today we're going to talk about smart
19:18
so today we're going to talk about smart spaces so
19:19
spaces so
19:19
spaces so the digital twins that we're going to
19:21
the digital twins that we're going to
19:21
the digital twins that we're going to talk about are for buildings and spaces
19:23
talk about are for buildings and spaces
19:23
talk about are for buildings and spaces so the data may be iot data from sensors
19:26
so the data may be iot data from sensors
19:26
so the data may be iot data from sensors in rooms or spaces
19:28
in rooms or spaces
19:28
in rooms or spaces it may be data about the room itself the
19:30
it may be data about the room itself the
19:30
it may be data about the room itself the dimensions
19:31
dimensions
19:31
dimensions how many people can fit in there what
19:33
how many people can fit in there what
19:33
how many people can fit in there what sort of furniture it's got there
19:35
sort of furniture it's got there
19:35
sort of furniture it's got there the digital twin also includes a whole
19:36
the digital twin also includes a whole
19:36
the digital twin also includes a whole host of analytics and algorithms
19:39
host of analytics and algorithms
19:39
host of analytics and algorithms so all sorts of artificial intelligence
19:41
so all sorts of artificial intelligence
19:41
so all sorts of artificial intelligence uh going on there that's probably a
19:42
uh going on there that's probably a
19:42
uh going on there that's probably a whole another
19:43
whole another
19:43
whole another whole another talk about what we can do
19:44
whole another talk about what we can do
19:44
whole another talk about what we can do specifically with smart spaces
19:46
specifically with smart spaces
19:46
specifically with smart spaces data from the digital twin and then also
19:49
data from the digital twin and then also
19:49
data from the digital twin and then also crucially we have that control element
19:50
crucially we have that control element
19:50
crucially we have that control element so we're able to make decisions
19:52
so we're able to make decisions
19:52
so we're able to make decisions and actually actuate those as well so
19:55
and actually actuate those as well so
19:55
and actually actuate those as well so you know in a smart building example it
19:57
you know in a smart building example it
19:57
you know in a smart building example it might be this room's too cold let's turn
19:59
might be this room's too cold let's turn
19:59
might be this room's too cold let's turn on the heaters and we can actually make
20:00
on the heaters and we can actually make
20:00
on the heaters and we can actually make those decisions
20:01
those decisions
20:01
those decisions within the digital twin so why would we
20:05
within the digital twin so why would we
20:05
within the digital twin so why would we use one over
20:06
use one over
20:06
use one over other data models and other ways of
20:07
other data models and other ways of
20:07
other data models and other ways of doing things um
20:09
doing things um
20:09
doing things um we really want to use them when we're
20:11
we really want to use them when we're
20:11
we really want to use them when we're closely aligning a physical and digital
20:12
closely aligning a physical and digital
20:12
closely aligning a physical and digital world so when
20:13
world so when
20:13
world so when um the digital world needs to represent
20:15
um the digital world needs to represent
20:15
um the digital world needs to represent the physical world
20:17
the physical world
20:17
the physical world in a very close fashion or maybe the
20:20
in a very close fashion or maybe the
20:20
in a very close fashion or maybe the you know needs to be almost real-time
20:22
you know needs to be almost real-time
20:22
you know needs to be almost real-time and live data we may not need it in a
20:24
and live data we may not need it in a
20:24
and live data we may not need it in a situation
20:25
situation
20:25
situation where there's a static physical object
20:27
where there's a static physical object
20:27
where there's a static physical object that doesn't change a lot or we don't
20:29
that doesn't change a lot or we don't
20:29
that doesn't change a lot or we don't really
20:29
really
20:29
really um we want to see it over time or we
20:31
um we want to see it over time or we
20:31
um we want to see it over time or we don't want to see it necessarily in its
20:33
don't want to see it necessarily in its
20:33
don't want to see it necessarily in its live state
20:34
live state
20:34
live state the other reason is because it helps
20:35
the other reason is because it helps
20:35
the other reason is because it helps collaboration between stakeholders
20:37
collaboration between stakeholders
20:37
collaboration between stakeholders especially
20:37
especially
20:37
especially especially with difficult and large
20:40
especially with difficult and large
20:40
especially with difficult and large problems
20:41
problems
20:41
problems we often have lots of different
20:42
we often have lots of different
20:42
we often have lots of different stakeholders involved and you can
20:44
stakeholders involved and you can
20:44
stakeholders involved and you can imagine with a smart building or a smart
20:45
imagine with a smart building or a smart
20:45
imagine with a smart building or a smart space
20:46
space
20:46
space there's lots of different people so
20:47
there's lots of different people so
20:47
there's lots of different people so there's you there might be people
20:49
there's you there might be people
20:49
there's you there might be people trying to book space there might be
20:51
trying to book space there might be
20:51
trying to book space there might be people
20:53
people
20:53
people trying to control the atmosphere the
20:54
trying to control the atmosphere the
20:54
trying to control the atmosphere the temperature
20:56
temperature
20:56
temperature um there might be other people more
20:58
um there might be other people more
20:58
um there might be other people more interested in how many people are using
20:59
interested in how many people are using
21:00
interested in how many people are using this space is social distancing be
21:01
this space is social distancing be
21:01
this space is social distancing be being adhered to lots of different
21:03
being adhered to lots of different
21:03
being adhered to lots of different stakeholders
21:05
stakeholders
21:05
stakeholders and also it just helps us integrate and
21:06
and also it just helps us integrate and
21:06
and also it just helps us integrate and abstract those underlying systems so
21:08
abstract those underlying systems so
21:08
abstract those underlying systems so really
21:09
really
21:09
really taking all of those uh different systems
21:12
taking all of those uh different systems
21:12
taking all of those uh different systems different building systems
21:13
different building systems
21:13
different building systems and integrating them together in one
21:16
and integrating them together in one
21:16
and integrating them together in one coherent place where we can get the most
21:18
coherent place where we can get the most
21:18
coherent place where we can get the most out of all of that data
21:20
out of all of that data
21:20
out of all of that data next i'm going to talk briefly about
21:22
next i'm going to talk briefly about
21:22
next i'm going to talk briefly about what is xr
21:24
what is xr
21:24
what is xr so we all know that we throw a lot of
21:26
so we all know that we throw a lot of
21:26
so we all know that we throw a lot of algorithms around
21:28
algorithms around
21:28
algorithms around in our daily life uh algorithm sorry
21:31
in our daily life uh algorithm sorry
21:31
in our daily life uh algorithm sorry acronyms
21:32
acronyms
21:32
acronyms uh so i just wanted to demystify a
21:34
uh so i just wanted to demystify a
21:34
uh so i just wanted to demystify a couple of these that i might be using
21:36
couple of these that i might be using
21:36
couple of these that i might be using so what's the difference between
21:38
so what's the difference between
21:38
so what's the difference between augmented virtual and mixed reality
21:41
augmented virtual and mixed reality
21:41
augmented virtual and mixed reality so augmented reality the version the
21:43
so augmented reality the version the
21:43
so augmented reality the version the example we have here is pokemon go so
21:45
example we have here is pokemon go so
21:45
example we have here is pokemon go so some of you
21:46
some of you
21:46
some of you might be familiar with pokemon go it's
21:48
might be familiar with pokemon go it's
21:48
might be familiar with pokemon go it's an app on your phone
21:49
an app on your phone
21:49
an app on your phone a game where you can use the camera to
21:52
a game where you can use the camera to
21:52
a game where you can use the camera to see pokemon
21:53
see pokemon
21:53
see pokemon appear in your real environment so it's
21:56
appear in your real environment so it's
21:56
appear in your real environment so it's using the real
21:57
using the real
21:57
using the real physical environment around you live and
21:59
physical environment around you live and
21:59
physical environment around you live and adding to it
22:01
adding to it
22:01
adding to it virtual reality is an entirely virtual
22:05
virtual reality is an entirely virtual
22:06
virtual reality is an entirely virtual environment that we place ourselves into
22:08
environment that we place ourselves into
22:08
environment that we place ourselves into so we're not looking at anything
22:10
so we're not looking at anything
22:10
so we're not looking at anything real when we're looking when we're in vr
22:12
real when we're looking when we're in vr
22:12
real when we're looking when we're in vr we're looking at entirely generated
22:14
we're looking at entirely generated
22:14
we're looking at entirely generated content
22:15
content
22:15
content and then there's mr or mixed reality
22:18
and then there's mr or mixed reality
22:18
and then there's mr or mixed reality which is a kind of a combination of the
22:19
which is a kind of a combination of the
22:19
which is a kind of a combination of the first two so we're using elements
22:21
first two so we're using elements
22:21
first two so we're using elements of real objects we're using
22:25
of real objects we're using
22:25
of real objects we're using elements of virtual objects and then
22:27
elements of virtual objects and then
22:27
elements of virtual objects and then interacting them together so
22:28
interacting them together so
22:28
interacting them together so really interesting we
22:31
really interesting we
22:31
really interesting we use the term xr or extended reality to
22:35
use the term xr or extended reality to
22:35
use the term xr or extended reality to actually come as a combination of these
22:37
actually come as a combination of these
22:37
actually come as a combination of these technologies so we may when we say xr we
22:39
technologies so we may when we say xr we
22:39
technologies so we may when we say xr we may actually be talking about
22:40
may actually be talking about
22:40
may actually be talking about ar vr or mixed reality
22:45
ar vr or mixed reality
22:45
ar vr or mixed reality and this piece of research comes into
22:47
and this piece of research comes into
22:47
and this piece of research comes into what we call
22:48
what we call
22:48
what we call avanade our experiences without uh
22:50
avanade our experiences without uh
22:50
avanade our experiences without uh boundaries research
22:51
boundaries research
22:51
boundaries research so you can go to the website and look
22:53
so you can go to the website and look
22:53
so you can go to the website and look this up and have a have a further read
22:55
this up and have a have a further read
22:55
this up and have a have a further read if you are interested
22:58
so what are the benefits of using xr
23:01
so what are the benefits of using xr
23:01
so what are the benefits of using xr um first one is one that comes up quite
23:03
um first one is one that comes up quite
23:04
um first one is one that comes up quite a lot is
23:04
a lot is
23:04
a lot is immersive learning so having people gain
23:07
immersive learning so having people gain
23:07
immersive learning so having people gain skills
23:08
skills
23:08
skills um in a sort of realistic setting
23:12
um in a sort of realistic setting
23:12
um in a sort of realistic setting without necessarily using expensive
23:14
without necessarily using expensive
23:14
without necessarily using expensive equipment or even dangerous equipment so
23:16
equipment or even dangerous equipment so
23:16
equipment or even dangerous equipment so safety is another huge factor we see
23:18
safety is another huge factor we see
23:18
safety is another huge factor we see a lot of applications of xr in
23:20
a lot of applications of xr in
23:20
a lot of applications of xr in manufacturing in mining
23:23
manufacturing in mining
23:23
manufacturing in mining in dangerous and high risk scenarios so
23:25
in dangerous and high risk scenarios so
23:25
in dangerous and high risk scenarios so that we can
23:26
that we can
23:26
that we can mitigate the data the the danger away
23:29
mitigate the data the the danger away
23:29
mitigate the data the the danger away from those users who may be doing
23:30
from those users who may be doing
23:30
from those users who may be doing something for the first time who may be
23:31
something for the first time who may be
23:32
something for the first time who may be still in that immersive learning stage
23:33
still in that immersive learning stage
23:33
still in that immersive learning stage and make that safer for them also we can
23:36
and make that safer for them also we can
23:36
and make that safer for them also we can personalize these spaces
23:37
personalize these spaces
23:37
personalize these spaces to uh you know use
23:40
to uh you know use
23:40
to uh you know use whatever we whatever we want so it's
23:42
whatever we whatever we want so it's
23:42
whatever we whatever we want so it's much more personalizable
23:45
much more personalizable
23:45
much more personalizable microsoft's really been at the forefront
23:47
microsoft's really been at the forefront
23:47
microsoft's really been at the forefront of some of this technology this is an
23:49
of some of this technology this is an
23:49
of some of this technology this is an example
23:50
example
23:50
example of the microsoft's hololens
23:54
of the microsoft's hololens
23:54
of the microsoft's hololens being used to allow a father and a
23:56
being used to allow a father and a
23:56
being used to allow a father and a daughter to interact in the same virtual
23:58
daughter to interact in the same virtual
23:58
daughter to interact in the same virtual space even though they're in physical
23:59
space even though they're in physical
24:00
space even though they're in physical spaces
24:00
spaces
24:00
spaces separate physical spaces so you know
24:03
separate physical spaces so you know
24:03
separate physical spaces so you know they can interact together they can
24:04
they can interact together they can
24:04
they can interact together they can see each other hear each other and
24:07
see each other hear each other and
24:07
see each other hear each other and uh in the same physical in the same
24:09
uh in the same physical in the same
24:09
uh in the same physical in the same virtual space even though they're in
24:11
virtual space even though they're in
24:11
virtual space even though they're in different physical spaces so the
24:14
different physical spaces so the
24:14
different physical spaces so the question
24:15
question
24:15
question we kind of asked ourselves and the story
24:17
we kind of asked ourselves and the story
24:17
we kind of asked ourselves and the story that i want to tell you through
24:18
that i want to tell you through
24:18
that i want to tell you through a series of short demos and showing you
24:21
a series of short demos and showing you
24:21
a series of short demos and showing you what we've been building
24:22
what we've been building
24:22
what we've been building is how can we use extended reality to
24:25
is how can we use extended reality to
24:25
is how can we use extended reality to connect distributed teams
24:26
connect distributed teams
24:26
connect distributed teams and the form of physical workplaces so a
24:29
and the form of physical workplaces so a
24:29
and the form of physical workplaces so a lot of us
24:29
lot of us
24:30
lot of us are now working in different places than
24:31
are now working in different places than
24:31
are now working in different places than we we thought we would be at the
24:33
we we thought we would be at the
24:33
we we thought we would be at the beginning of the year
24:34
beginning of the year
24:34
beginning of the year we're working at home we're working
24:37
we're working at home we're working
24:37
we're working at home we're working uh with people all across the globe
24:40
uh with people all across the globe
24:40
uh with people all across the globe potentially
24:41
potentially
24:41
potentially in our organizations and we're outside
24:43
in our organizations and we're outside
24:43
in our organizations and we're outside of those former
24:44
of those former
24:44
of those former former workplaces so the idea behind
24:46
former workplaces so the idea behind
24:46
former workplaces so the idea behind this was how do we
24:48
this was how do we
24:48
this was how do we use some of this technology to go back
24:51
use some of this technology to go back
24:51
use some of this technology to go back to
24:52
to
24:52
to not go back to the office but look at
24:54
not go back to the office but look at
24:54
not go back to the office but look at different ways
24:55
different ways
24:55
different ways of different ways of thinking about
25:00
of different ways of thinking about
25:00
of different ways of thinking about using physical spaces again
25:08
so um i wanted to go through some demos
25:10
so um i wanted to go through some demos
25:10
so um i wanted to go through some demos and i'm going to kind of walk you
25:11
and i'm going to kind of walk you
25:11
and i'm going to kind of walk you through the process
25:12
through the process
25:12
through the process of how we get from digital twins to
25:14
of how we get from digital twins to
25:14
of how we get from digital twins to extended reality and everything that's
25:16
extended reality and everything that's
25:16
extended reality and everything that's involved
25:18
involved
25:18
involved so the first thing i want to show you is
25:20
so the first thing i want to show you is
25:20
so the first thing i want to show you is actually
25:21
actually
25:21
actually not that is this so this
25:24
not that is this so this
25:24
not that is this so this is a very complicated architectural
25:27
is a very complicated architectural
25:27
is a very complicated architectural diagram
25:28
diagram
25:28
diagram of a physical space so this is an
25:31
of a physical space so this is an
25:31
of a physical space so this is an accenture building that's still being
25:33
accenture building that's still being
25:33
accenture building that's still being built in manhattan
25:35
built in manhattan
25:36
built in manhattan in midtown in manhattan and this is an
25:38
in midtown in manhattan and this is an
25:38
in midtown in manhattan and this is an architectural diagram
25:40
architectural diagram
25:40
architectural diagram so this contains all of the information
25:42
so this contains all of the information
25:42
so this contains all of the information about the physical space in terms of
25:44
about the physical space in terms of
25:44
about the physical space in terms of the layout the furniture what facilities
25:47
the layout the furniture what facilities
25:47
the layout the furniture what facilities are available so you can see here we
25:48
are available so you can see here we
25:48
are available so you can see here we have
25:49
have
25:49
have a coffee point a kitchenette
25:52
a coffee point a kitchenette
25:52
a coffee point a kitchenette restrooms a reception area all sorts of
25:55
restrooms a reception area all sorts of
25:55
restrooms a reception area all sorts of different information
25:56
different information
25:56
different information but this is useful for if you were in
25:59
but this is useful for if you were in
25:59
but this is useful for if you were in construction putting the building
26:00
construction putting the building
26:00
construction putting the building together
26:02
together
26:02
together but that's about it so what microsoft
26:04
but that's about it so what microsoft
26:04
but that's about it so what microsoft have been working on and we have been
26:06
have been working on and we have been
26:06
have been working on and we have been utilizing avenatt is the azure indoor
26:10
utilizing avenatt is the azure indoor
26:10
utilizing avenatt is the azure indoor maps
26:12
maps
26:12
maps so indoor maps is a new part of bing
26:15
so indoor maps is a new part of bing
26:16
so indoor maps is a new part of bing maps soon to be agile maps
26:18
maps soon to be agile maps
26:18
maps soon to be agile maps which allows you to map and generate
26:21
which allows you to map and generate
26:21
which allows you to map and generate your own
26:21
your own
26:21
your own indoor spaces as part of bing maps so
26:24
indoor spaces as part of bing maps so
26:24
indoor spaces as part of bing maps so this could be a campus a single building
26:27
this could be a campus a single building
26:27
this could be a campus a single building a section of a building anything you'd
26:28
a section of a building anything you'd
26:28
a section of a building anything you'd like and you can see here that
26:30
like and you can see here that
26:30
like and you can see here that we've actually used algorithms provided
26:34
we've actually used algorithms provided
26:34
we've actually used algorithms provided by microsoft to go from that
26:35
by microsoft to go from that
26:35
by microsoft to go from that quite technical architectural drawing to
26:38
quite technical architectural drawing to
26:38
quite technical architectural drawing to this quite nice
26:39
this quite nice
26:39
this quite nice familiar looking map so we've taken off
26:42
familiar looking map so we've taken off
26:42
familiar looking map so we've taken off a lot of the excess
26:43
a lot of the excess
26:43
a lot of the excess information and we've overlaid our own
26:45
information and we've overlaid our own
26:45
information and we've overlaid our own using microsoft's apis
26:48
using microsoft's apis
26:48
using microsoft's apis so in this case we're looking at
26:51
so in this case we're looking at
26:51
so in this case we're looking at occupancy so we have iot sensors on
26:54
occupancy so we have iot sensors on
26:54
occupancy so we have iot sensors on every desk and in every meeting room we
26:56
every desk and in every meeting room we
26:56
every desk and in every meeting room we have
26:57
have
26:57
have information in the digital twin about
26:59
information in the digital twin about
26:59
information in the digital twin about how the rooms have been booked
27:01
how the rooms have been booked
27:01
how the rooms have been booked we also have those sensors which allow
27:02
we also have those sensors which allow
27:02
we also have those sensors which allow us to paint a picture of how the space
27:04
us to paint a picture of how the space
27:04
us to paint a picture of how the space is being used
27:05
is being used
27:05
is being used so this is just some sample data because
27:07
so this is just some sample data because
27:07
so this is just some sample data because this building hasn't been built yet but
27:08
this building hasn't been built yet but
27:08
this building hasn't been built yet but for example in this bank of desks i can
27:10
for example in this bank of desks i can
27:10
for example in this bank of desks i can see that three of those were actually
27:12
see that three of those were actually
27:12
see that three of those were actually booked out and in use this orange one
27:14
booked out and in use this orange one
27:14
booked out and in use this orange one here might be
27:16
here might be
27:16
here might be booked out but it is somebody's actually
27:18
booked out but it is somebody's actually
27:18
booked out but it is somebody's actually sat there
27:19
sat there
27:19
sat there but it's not booked out and then the
27:20
but it's not booked out and then the
27:20
but it's not booked out and then the green ones may be okay that
27:22
green ones may be okay that
27:22
green ones may be okay that that room is actually there's nobody sat
27:24
that room is actually there's nobody sat
27:24
that room is actually there's nobody sat there at the moment the sensor's coming
27:25
there at the moment the sensor's coming
27:25
there at the moment the sensor's coming back saying it's empty and it hasn't
27:26
back saying it's empty and it hasn't
27:26
back saying it's empty and it hasn't been booked so this is immediately
27:28
been booked so this is immediately
27:28
been booked so this is immediately available
27:28
available
27:28
available and i can look at this and very quickly
27:30
and i can look at this and very quickly
27:30
and i can look at this and very quickly say okay these are the desks i can
27:32
say okay these are the desks i can
27:32
say okay these are the desks i can are available um this is the meeting
27:34
are available um this is the meeting
27:34
are available um this is the meeting rooms that are available
27:36
rooms that are available
27:36
rooms that are available and then we can start doing other things
27:37
and then we can start doing other things
27:37
and then we can start doing other things like uh wayfinding so how do i get from
27:40
like uh wayfinding so how do i get from
27:40
like uh wayfinding so how do i get from reception maybe i need to go up a floor
27:42
reception maybe i need to go up a floor
27:42
reception maybe i need to go up a floor maybe i need to go down a floor uh to
27:45
maybe i need to go down a floor uh to
27:45
maybe i need to go down a floor uh to find the space
27:45
find the space
27:45
find the space that i'm looking at and the i this idea
27:49
that i'm looking at and the i this idea
27:49
that i'm looking at and the i this idea then
27:50
then
27:50
then led to a discussion on visualizations
27:53
led to a discussion on visualizations
27:53
led to a discussion on visualizations so this is a very sort of familiar way
27:56
so this is a very sort of familiar way
27:56
so this is a very sort of familiar way of visualizing
27:57
of visualizing
27:57
of visualizing visualizing a physical space
28:00
visualizing a physical space
28:00
visualizing a physical space you know most of us who are familiar
28:02
you know most of us who are familiar
28:02
you know most of us who are familiar with looking at maps regularly will look
28:04
with looking at maps regularly will look
28:04
with looking at maps regularly will look at this and say okay i kind of
28:05
at this and say okay i kind of
28:06
at this and say okay i kind of understand
28:06
understand
28:06
understand what's a wall what's a desk what's a
28:09
what's a wall what's a desk what's a
28:09
what's a wall what's a desk what's a piece of glass
28:10
piece of glass
28:10
piece of glass but then we started asking ourselves in
28:11
but then we started asking ourselves in
28:11
but then we started asking ourselves in this particular example
28:13
this particular example
28:13
this particular example this building hasn't actually been built
28:14
this building hasn't actually been built
28:14
this building hasn't actually been built yet so
28:16
yet so
28:16
yet so what is it actually going to look like
28:18
what is it actually going to look like
28:18
what is it actually going to look like how can we make users more familiar with
28:20
how can we make users more familiar with
28:20
how can we make users more familiar with the space
28:21
the space
28:21
the space if they have never been there before and
28:23
if they have never been there before and
28:23
if they have never been there before and they don't know what it looks like
28:25
they don't know what it looks like
28:25
they don't know what it looks like so then we started thinking about
28:27
so then we started thinking about
28:27
so then we started thinking about changing this from a 2d perspective into
28:29
changing this from a 2d perspective into
28:29
changing this from a 2d perspective into a 3d map so maybe we could get more
28:31
a 3d map so maybe we could get more
28:31
a 3d map so maybe we could get more information
28:32
information
28:32
information about the layout of the building so what
28:36
about the layout of the building so what
28:36
about the layout of the building so what we
28:36
we
28:36
we were able to do using the apis available
28:39
were able to do using the apis available
28:39
were able to do using the apis available within
28:42
within
28:42
within azure maps was actually generate this 3d
28:44
azure maps was actually generate this 3d
28:44
azure maps was actually generate this 3d model of the building
28:46
model of the building
28:46
model of the building so this is the same information and
28:48
so this is the same information and
28:48
so this is the same information and we're able to
28:49
we're able to
28:49
we're able to automatically generate from our map
28:51
automatically generate from our map
28:51
automatically generate from our map which is our sort of
28:52
which is our sort of
28:52
which is our sort of data source uh the same meeting rooms
28:56
data source uh the same meeting rooms
28:56
data source uh the same meeting rooms the locations of the desk we're able to
28:59
the locations of the desk we're able to
28:59
the locations of the desk we're able to import automatically
29:00
import automatically
29:00
import automatically a model of a desk and put it where
29:04
a model of a desk and put it where
29:04
a model of a desk and put it where the desks appear in real life for
29:06
the desks appear in real life for
29:06
the desks appear in real life for example we can see the meeting rooms
29:09
example we can see the meeting rooms
29:09
example we can see the meeting rooms we can even visualize
29:12
we can even visualize
29:12
we can even visualize where the glass walls are as opposed to
29:15
where the glass walls are as opposed to
29:15
where the glass walls are as opposed to the solid walls
29:16
the solid walls
29:16
the solid walls and so we can get a real idea of this
29:17
and so we can get a real idea of this
29:17
and so we can get a real idea of this space what it looks like
29:20
space what it looks like
29:20
space what it looks like how we might use it and that's all
29:22
how we might use it and that's all
29:22
how we might use it and that's all completely
29:23
completely
29:23
completely generated automatically just simply from
29:26
generated automatically just simply from
29:26
generated automatically just simply from that step process of going from the
29:28
that step process of going from the
29:28
that step process of going from the architectural diagrams
29:29
architectural diagrams
29:29
architectural diagrams to the 2d azure map
29:33
to the 2d azure map
29:33
to the 2d azure map to this 3d model but then okay this is
29:37
to this 3d model but then okay this is
29:37
to this 3d model but then okay this is okay
29:37
okay
29:38
okay we can see the walls i can see the desks
29:39
we can see the walls i can see the desks
29:40
we can see the walls i can see the desks i can see the
29:41
i can see the
29:41
i can see the the the units but uh and this is great
29:43
the the units but uh and this is great
29:43
the the units but uh and this is great for visualization for visualizing
29:46
for visualization for visualizing
29:46
for visualization for visualizing the desk occupancy so on this 3d map i
29:49
the desk occupancy so on this 3d map i
29:49
the desk occupancy so on this 3d map i can kind of get an understanding of the
29:50
can kind of get an understanding of the
29:50
can kind of get an understanding of the space
29:51
space
29:51
space and i can see which desks are occupied
29:53
and i can see which desks are occupied
29:53
and i can see which desks are occupied or not occupied
29:56
but then we need to stop then we started
29:58
but then we need to stop then we started
29:58
but then we need to stop then we started thinking well how can we then actually
29:59
thinking well how can we then actually
30:00
thinking well how can we then actually improve this
30:01
improve this
30:01
improve this so that we can walk around the space
30:03
so that we can walk around the space
30:03
so that we can walk around the space explore the space
30:04
explore the space
30:04
explore the space and manage to see what it's like and
30:07
and manage to see what it's like and
30:07
and manage to see what it's like and that also has
30:08
that also has
30:08
that also has you know this all has implications with
30:10
you know this all has implications with
30:10
you know this all has implications with the covid pandemic we can use this for
30:12
the covid pandemic we can use this for
30:12
the covid pandemic we can use this for physical distancing but we could also
30:13
physical distancing but we could also
30:13
physical distancing but we could also use it
30:14
use it
30:14
use it for actually linking up with our
30:16
for actually linking up with our
30:16
for actually linking up with our colleagues and using and interacting in
30:18
colleagues and using and interacting in
30:18
colleagues and using and interacting in digital spaces as well
30:20
digital spaces as well
30:20
digital spaces as well so after we got to this point we started
30:23
so after we got to this point we started
30:23
so after we got to this point we started experimenting
30:25
experimenting
30:25
experimenting with um that this is just sorry this is
30:28
with um that this is just sorry this is
30:28
with um that this is just sorry this is just an example of
30:30
just an example of
30:30
just an example of how we actually use the 2d information
30:32
how we actually use the 2d information
30:32
how we actually use the 2d information so we are
30:33
so we are
30:33
so we are looking at desk occupancy things like
30:35
looking at desk occupancy things like
30:35
looking at desk occupancy things like temperature air quality lots and lots of
30:37
temperature air quality lots and lots of
30:37
temperature air quality lots and lots of different pieces of data
30:39
different pieces of data
30:39
different pieces of data and this is how just an idea of how we
30:40
and this is how just an idea of how we
30:40
and this is how just an idea of how we how we would visualize that so in a
30:42
how we would visualize that so in a
30:42
how we would visualize that so in a traditional portal
30:43
traditional portal
30:43
traditional portal in a 2d sense but now we're kind of
30:46
in a 2d sense but now we're kind of
30:46
in a 2d sense but now we're kind of taking that
30:47
taking that
30:47
taking that to what do we do in a for a 3d
30:49
to what do we do in a for a 3d
30:50
to what do we do in a for a 3d environment or one that we can use
30:52
environment or one that we can use
30:52
environment or one that we can use so in exactly the same way um one of the
30:56
so in exactly the same way um one of the
30:56
so in exactly the same way um one of the one of the amazing things that agile
30:58
one of the amazing things that agile
30:58
one of the amazing things that agile maps will have the ability to do at some
31:00
maps will have the ability to do at some
31:00
maps will have the ability to do at some point in the future is actually
31:01
point in the future is actually
31:01
point in the future is actually not require those architectural diagrams
31:03
not require those architectural diagrams
31:03
not require those architectural diagrams to be able to go from images of floor
31:05
to be able to go from images of floor
31:05
to be able to go from images of floor maps
31:06
maps
31:06
maps uh using ai to generate the same 2d
31:09
uh using ai to generate the same 2d
31:09
uh using ai to generate the same 2d spaces
31:11
spaces
31:11
spaces on 2d representation of the spaces then
31:14
on 2d representation of the spaces then
31:14
on 2d representation of the spaces then we can actually go
31:15
we can actually go
31:15
we can actually go from the 2d to the 3d using tools like
31:17
from the 2d to the 3d using tools like
31:17
from the 2d to the 3d using tools like unity
31:18
unity
31:18
unity and then we can augment those once we
31:21
and then we can augment those once we
31:21
and then we can augment those once we have the layouts and the plans of the
31:22
have the layouts and the plans of the
31:22
have the layouts and the plans of the building
31:23
building
31:23
building with furniture assets colors and
31:26
with furniture assets colors and
31:26
with furniture assets colors and textures that we would expect to see in
31:27
textures that we would expect to see in
31:27
textures that we would expect to see in the building
31:28
the building
31:28
the building so for this example i'm actually going
31:30
so for this example i'm actually going
31:30
so for this example i'm actually going to change the building perspective and
31:31
to change the building perspective and
31:32
to change the building perspective and go through an example of
31:34
go through an example of
31:34
go through an example of our london office instead of the
31:37
our london office instead of the
31:37
our london office instead of the accenture new york
31:38
accenture new york
31:38
accenture new york office and this is just because i'm more
31:41
office and this is just because i'm more
31:41
office and this is just because i'm more familiar with this space
31:42
familiar with this space
31:42
familiar with this space so i'm currently sat in that digital
31:45
so i'm currently sat in that digital
31:46
so i'm currently sat in that digital space using a rendering
31:49
space using a rendering
31:49
space using a rendering because i know what it looks like and i
31:50
because i know what it looks like and i
31:50
because i know what it looks like and i could more easily model that effectively
31:53
could more easily model that effectively
31:53
could more easily model that effectively so here's a couple of images on the left
31:55
so here's a couple of images on the left
31:55
so here's a couple of images on the left you can see
31:57
you can see
31:57
you can see that there's a real image of the space
32:00
that there's a real image of the space
32:00
that there's a real image of the space versus the
32:01
versus the
32:01
versus the um virtual image of the space
32:04
um virtual image of the space
32:04
um virtual image of the space and this particular picture actually
32:07
and this particular picture actually
32:07
and this particular picture actually kicked off
32:07
kicked off
32:08
kicked off a um something for us which is once we
32:10
a um something for us which is once we
32:10
a um something for us which is once we had the walls and the desks
32:11
had the walls and the desks
32:11
had the walls and the desks automatically generated and i started
32:14
automatically generated and i started
32:14
automatically generated and i started adding
32:15
adding
32:15
adding personalized hand-built features like
32:17
personalized hand-built features like
32:17
personalized hand-built features like the windows
32:18
the windows
32:18
the windows and the stripe on the ceiling for
32:20
and the stripe on the ceiling for
32:20
and the stripe on the ceiling for example we then got the feedback that
32:22
example we then got the feedback that
32:22
example we then got the feedback that this space is
32:23
this space is
32:23
this space is very empty it feels unused it looks
32:27
very empty it feels unused it looks
32:27
very empty it feels unused it looks too clean the real offices are always
32:30
too clean the real offices are always
32:30
too clean the real offices are always you know cluttered with stationery
32:32
you know cluttered with stationery
32:32
you know cluttered with stationery laptops um
32:34
laptops um
32:34
laptops um people's uh you know personal effects as
32:36
people's uh you know personal effects as
32:36
people's uh you know personal effects as you can see on that image
32:37
you can see on that image
32:37
you can see on that image in the left so what we started doing was
32:40
in the left so what we started doing was
32:40
in the left so what we started doing was actually using procedural generation
32:42
actually using procedural generation
32:42
actually using procedural generation so we have imported a
32:45
so we have imported a
32:45
so we have imported a bunch of different assets like computer
32:47
bunch of different assets like computer
32:47
bunch of different assets like computer monitors keyboards
32:49
monitors keyboards
32:49
monitors keyboards um and you know desk clutter so there's
32:53
um and you know desk clutter so there's
32:53
um and you know desk clutter so there's some mugs
32:54
some mugs
32:54
some mugs and some coffee cups and we know where
32:57
and some coffee cups and we know where
32:57
and some coffee cups and we know where the desks are from the information in
32:59
the desks are from the information in
32:59
the desks are from the information in the 2d maps
33:00
the 2d maps
33:00
the 2d maps so we can add the sort of desk
33:03
so we can add the sort of desk
33:03
so we can add the sort of desk clutter and make it feel more lived in
33:05
clutter and make it feel more lived in
33:05
clutter and make it feel more lived in we can also add other
33:07
we can also add other
33:07
we can also add other pieces of furniture from that map once
33:09
pieces of furniture from that map once
33:09
pieces of furniture from that map once we have an idea and once we can
33:11
we have an idea and once we can
33:11
we have an idea and once we can import objects easily we can just sort
33:14
import objects easily we can just sort
33:14
import objects easily we can just sort of place them around
33:16
of place them around
33:16
of place them around to give a more accurate representation
33:17
to give a more accurate representation
33:17
to give a more accurate representation of that physical space
33:19
of that physical space
33:19
of that physical space so here's just another rendering of what
33:20
so here's just another rendering of what
33:20
so here's just another rendering of what that space looks like
33:22
that space looks like
33:22
that space looks like rendered in blender so it's quite shiny
33:25
rendered in blender so it's quite shiny
33:25
rendered in blender so it's quite shiny and polished and
33:25
and polished and
33:26
and polished and nice um what i want to show you next
33:29
nice um what i want to show you next
33:29
nice um what i want to show you next is actually a video of us using the
33:31
is actually a video of us using the
33:31
is actually a video of us using the space in
33:34
space in
33:34
space in in a oculus headset using a tool called
33:37
in a oculus headset using a tool called
33:37
in a oculus headset using a tool called alt space vr
33:38
alt space vr
33:38
alt space vr so this is the same space but rendered
33:41
so this is the same space but rendered
33:41
so this is the same space but rendered on
33:42
on
33:42
on a headset so it's a little bit lighter
33:43
a headset so it's a little bit lighter
33:43
a headset so it's a little bit lighter we've lost some of the nice textures but
33:45
we've lost some of the nice textures but
33:45
we've lost some of the nice textures but we still get a really good idea of space
33:47
we still get a really good idea of space
33:47
we still get a really good idea of space so in this case we can use
33:50
so in this case we can use
33:50
so in this case we can use the headset so i have one here uh we can
33:53
the headset so i have one here uh we can
33:53
the headset so i have one here uh we can wear those and we can explore the space
33:55
wear those and we can explore the space
33:55
wear those and we can explore the space in uh this sort of 3d
33:59
in uh this sort of 3d
33:59
in uh this sort of 3d digital the 3d digital environment and
34:02
digital the 3d digital environment and
34:02
digital the 3d digital environment and world
34:04
world
34:04
world and we can move around that space we can
34:06
and we can move around that space we can
34:06
and we can move around that space we can collaborate
34:08
collaborate
34:08
collaborate with other people so we can also install
34:11
with other people so we can also install
34:11
with other people so we can also install bespoke tools into those spaces so for
34:13
bespoke tools into those spaces so for
34:13
bespoke tools into those spaces so for example here we're watching
34:15
example here we're watching
34:15
example here we're watching a video presentation with my colleague
34:17
a video presentation with my colleague
34:17
a video presentation with my colleague christy whose avatar is there
34:19
christy whose avatar is there
34:19
christy whose avatar is there we can also fiddle around with
34:22
we can also fiddle around with
34:22
we can also fiddle around with interactable objects in those spaces so
34:24
interactable objects in those spaces so
34:24
interactable objects in those spaces so this is us trying to play
34:25
this is us trying to play
34:25
this is us trying to play very badly catch because neither of us
34:28
very badly catch because neither of us
34:28
very badly catch because neither of us are very coordinated
34:30
are very coordinated
34:30
are very coordinated and we can use the space in sort of any
34:32
and we can use the space in sort of any
34:32
and we can use the space in sort of any way we we want
34:33
way we we want
34:33
way we we want in that sort of interactable way
34:36
in that sort of interactable way
34:36
in that sort of interactable way we can also use custom um
34:39
we can also use custom um
34:39
we can also use custom um pieces of code so this is us playing a
34:41
pieces of code so this is us playing a
34:41
pieces of code so this is us playing a quiz together
34:42
quiz together
34:42
quiz together take selfies lots of you know fun things
34:45
take selfies lots of you know fun things
34:45
take selfies lots of you know fun things to do in vr
34:46
to do in vr
34:46
to do in vr so it does seem like uh some fun on the
34:49
so it does seem like uh some fun on the
34:49
so it does seem like uh some fun on the surface but actually there's very real
34:51
surface but actually there's very real
34:51
surface but actually there's very real applications to this
34:52
applications to this
34:52
applications to this you know remote um working remotely
34:56
you know remote um working remotely
34:56
you know remote um working remotely together
34:57
together
34:57
together we can do sort of design thinking
34:59
we can do sort of design thinking
34:59
we can do sort of design thinking workshops we can do lots of the the
35:02
workshops we can do lots of the the
35:02
workshops we can do lots of the the tasks that are difficult to do
35:04
tasks that are difficult to do
35:04
tasks that are difficult to do uh around 3d modeling sharing
35:06
uh around 3d modeling sharing
35:06
uh around 3d modeling sharing information on 3d objects
35:10
information on 3d objects
35:10
information on 3d objects and really exploring different uh
35:12
and really exploring different uh
35:12
and really exploring different uh techniques we can also do virtual tours
35:13
techniques we can also do virtual tours
35:14
techniques we can also do virtual tours so this is actually
35:14
so this is actually
35:14
so this is actually me showing my colleague christy who's
35:16
me showing my colleague christy who's
35:16
me showing my colleague christy who's from the seattle office and has ever
35:17
from the seattle office and has ever
35:17
from the seattle office and has ever been to the london office
35:19
been to the london office
35:19
been to the london office showing her the boardroom for example of
35:21
showing her the boardroom for example of
35:21
showing her the boardroom for example of that chemistry office
35:24
that chemistry office
35:24
that chemistry office so i wanted to use an example
35:27
so i wanted to use an example
35:27
so i wanted to use an example as well of why ai is so important and
35:31
as well of why ai is so important and
35:31
as well of why ai is so important and why
35:31
why
35:31
why procedural generation is so important
35:33
procedural generation is so important
35:33
procedural generation is so important when we're thinking about virtual spaces
35:35
when we're thinking about virtual spaces
35:35
when we're thinking about virtual spaces uh the the id the example that jumped
35:39
uh the the id the example that jumped
35:39
uh the the id the example that jumped into my mind
35:40
into my mind
35:40
into my mind was to look at all of the work that
35:44
was to look at all of the work that
35:44
was to look at all of the work that black shark ai have done with microsoft
35:46
black shark ai have done with microsoft
35:46
black shark ai have done with microsoft flight simulator it's a really good
35:47
flight simulator it's a really good
35:47
flight simulator it's a really good example of
35:49
example of
35:49
example of procedural generation on an absolutely
35:51
procedural generation on an absolutely
35:51
procedural generation on an absolutely enormous scale
35:52
enormous scale
35:52
enormous scale so looking here this is actually if i
35:54
so looking here this is actually if i
35:54
so looking here this is actually if i can use my
35:55
can use my
35:55
can use my pointer um on the left we can actually
35:58
pointer um on the left we can actually
35:58
pointer um on the left we can actually see some paul's cathedral in london and
36:00
see some paul's cathedral in london and
36:00
see some paul's cathedral in london and the
36:00
the
36:00
the um that's not that clear but maybe i can
36:03
um that's not that clear but maybe i can
36:03
um that's not that clear but maybe i can use pen
36:07
no okay sorry um so that's actually the
36:11
no okay sorry um so that's actually the
36:11
no okay sorry um so that's actually the the building microsoft had no
36:13
the building microsoft had no
36:13
the building microsoft had no information about what the building
36:14
information about what the building
36:14
information about what the building looks like other than its shape as seen
36:16
looks like other than its shape as seen
36:16
looks like other than its shape as seen from space
36:17
from space
36:17
from space but we've managed to create these really
36:18
but we've managed to create these really
36:18
but we've managed to create these really realistic um
36:20
realistic um
36:20
realistic um really realistically looking
36:21
really realistically looking
36:21
really realistically looking environments for the flight simulator
36:23
environments for the flight simulator
36:23
environments for the flight simulator platform
36:23
platform
36:23
platform and on the right we have the one
36:25
and on the right we have the one
36:25
and on the right we have the one manhattan west location so that's the
36:27
manhattan west location so that's the
36:27
manhattan west location so that's the first building i showed you
36:28
first building i showed you
36:28
first building i showed you but you can see it has it hasn't been
36:30
but you can see it has it hasn't been
36:30
but you can see it has it hasn't been built there yet in that photo
36:31
built there yet in that photo
36:31
built there yet in that photo so it's still that just there that
36:35
so it's still that just there that
36:35
so it's still that just there that that ground piece but what we can do is
36:37
that ground piece but what we can do is
36:37
that ground piece but what we can do is using all this ai and procedure
36:38
using all this ai and procedure
36:38
using all this ai and procedure generation is make these amazing
36:40
generation is make these amazing
36:40
generation is make these amazing environments um with you know quite a
36:42
environments um with you know quite a
36:42
environments um with you know quite a lot of time and effort
36:43
lot of time and effort
36:43
lot of time and effort so black shark spent a lot of
36:46
so black shark spent a lot of
36:46
so black shark spent a lot of uh time and a lot of there was a lot of
36:49
uh time and a lot of there was a lot of
36:49
uh time and a lot of there was a lot of investment in this absolutely massive
36:51
investment in this absolutely massive
36:51
investment in this absolutely massive project to 3d model the entire world
36:53
project to 3d model the entire world
36:53
project to 3d model the entire world but the key point for me on this is that
36:55
but the key point for me on this is that
36:55
but the key point for me on this is that it's actually
36:57
it's actually
36:57
it's actually some uh procedural generation
37:00
some uh procedural generation
37:00
some uh procedural generation so for example the building the
37:02
so for example the building the
37:02
so for example the building the buildings on the left are just generic
37:04
buildings on the left are just generic
37:04
buildings on the left are just generic buildings apart from paul's cathedral
37:06
buildings apart from paul's cathedral
37:06
buildings apart from paul's cathedral and some bespoke handmade items to bring
37:09
and some bespoke handmade items to bring
37:09
and some bespoke handmade items to bring that realism
37:10
that realism
37:10
that realism so some paul's cathedral has been hand
37:11
so some paul's cathedral has been hand
37:11
so some paul's cathedral has been hand modeled by somebody um
37:13
modeled by somebody um
37:13
modeled by somebody um uh some of the buildings on the new york
37:15
uh some of the buildings on the new york
37:15
uh some of the buildings on the new york skyline have been hand modeled
37:17
skyline have been hand modeled
37:17
skyline have been hand modeled so it's really a blend of using the ai
37:20
so it's really a blend of using the ai
37:20
so it's really a blend of using the ai the procedural generation the automatic
37:22
the procedural generation the automatic
37:22
the procedural generation the automatic techniques and a little bit of
37:23
techniques and a little bit of
37:23
techniques and a little bit of human creativity and it to be honest
37:26
human creativity and it to be honest
37:26
human creativity and it to be honest it's a bit of a labor
37:27
it's a bit of a labor
37:27
it's a bit of a labor labor of love as well to create some of
37:28
labor of love as well to create some of
37:28
labor of love as well to create some of these 3d digital spaces
37:31
these 3d digital spaces
37:31
these 3d digital spaces um so with that i will say thank you and
37:34
um so with that i will say thank you and
37:34
um so with that i will say thank you and i'll hand back to
37:35
i'll hand back to
37:35
i'll hand back to alicia for the panel session
37:41
cool thank you yeah that was interesting
37:46
cool thank you yeah that was interesting
37:46
cool thank you yeah that was interesting i i i never realized that the stuff that
37:50
i i i never realized that the stuff that
37:50
i i i never realized that the stuff that they used in flight simulator is
37:51
they used in flight simulator is
37:52
they used in flight simulator is actually
37:52
actually
37:52
actually usable on inside spaces as well so
37:55
usable on inside spaces as well so
37:55
usable on inside spaces as well so that's cool
37:57
that's cool
37:57
that's cool huh yeah definitely all the techniques
37:59
huh yeah definitely all the techniques
37:59
huh yeah definitely all the techniques and all of the
38:00
and all of the
38:00
and all of the the ai and all the procedural generation
38:03
the ai and all the procedural generation
38:03
the ai and all the procedural generation is
38:03
is
38:03
is as long as we're stuck it's all for me
38:05
as long as we're stuck it's all for me
38:05
as long as we're stuck it's all for me it's all about the data right
38:07
it's all about the data right
38:07
it's all about the data right so they started from uh having big maps
38:10
so they started from uh having big maps
38:10
so they started from uh having big maps which is a huge asset
38:12
which is a huge asset
38:12
which is a huge asset for microsoft to be able to generate
38:13
for microsoft to be able to generate
38:13
for microsoft to be able to generate those 3d spaces we're starting from
38:15
those 3d spaces we're starting from
38:15
those 3d spaces we're starting from either architectural diagrams
38:16
either architectural diagrams
38:16
either architectural diagrams or image layouts that build it
38:20
or image layouts that build it
38:20
or image layouts that build it yeah cool um so we do have a question
38:24
yeah cool um so we do have a question
38:24
yeah cool um so we do have a question from the audience you've shown a lot of
38:25
from the audience you've shown a lot of
38:25
from the audience you've shown a lot of stuff actually you've combined iot
38:28
stuff actually you've combined iot
38:28
stuff actually you've combined iot with virtual reality um
38:31
with virtual reality um
38:32
with virtual reality um there's definitely a lot of software
38:34
there's definitely a lot of software
38:34
there's definitely a lot of software engineering going into this
38:36
engineering going into this
38:36
engineering going into this um and then there's the ai
38:39
um and then there's the ai
38:39
um and then there's the ai i guess for for the object generate
38:41
i guess for for the object generate
38:41
i guess for for the object generate generation
38:42
generation
38:42
generation um how can you explain a bit more
38:47
um how can you explain a bit more
38:47
um how can you explain a bit more how the procedure generation works uh
38:49
how the procedure generation works uh
38:50
how the procedure generation works uh especially since you've also got these
38:52
especially since you've also got these
38:52
especially since you've also got these fixed assets in your scene yeah so
38:55
fixed assets in your scene yeah so
38:55
fixed assets in your scene yeah so it's kind of a combination of techniques
38:57
it's kind of a combination of techniques
38:57
it's kind of a combination of techniques um really
38:58
um really
38:58
um really the there's a lot of artificial
39:00
the there's a lot of artificial
39:00
the there's a lot of artificial intelligence and going from
39:02
intelligence and going from
39:02
intelligence and going from that data product which is quite complex
39:04
that data product which is quite complex
39:04
that data product which is quite complex unstructured data
39:05
unstructured data
39:06
unstructured data to something that we can actually
39:07
to something that we can actually
39:08
to something that we can actually understand in
39:09
understand in
39:09
understand in sort of a structured data format um
39:12
sort of a structured data format um
39:12
sort of a structured data format um so when we go for example from the
39:14
so when we go for example from the
39:14
so when we go for example from the architectural diagram
39:15
architectural diagram
39:15
architectural diagram into the 2d azure map there's a lot of
39:19
into the 2d azure map there's a lot of
39:19
into the 2d azure map there's a lot of assumptions that have to be made
39:20
assumptions that have to be made
39:20
assumptions that have to be made by ai systems because about what
39:23
by ai systems because about what
39:24
by ai systems because about what certain objects are where certain rooms
39:26
certain objects are where certain rooms
39:26
certain objects are where certain rooms will go especially if we're looking at
39:28
will go especially if we're looking at
39:28
will go especially if we're looking at using the future capability of going
39:30
using the future capability of going
39:30
using the future capability of going from a flat png
39:32
from a flat png
39:32
from a flat png image we have to actually work out what
39:34
image we have to actually work out what
39:34
image we have to actually work out what is in the image what are we actually
39:35
is in the image what are we actually
39:35
is in the image what are we actually looking at
39:36
looking at
39:36
looking at um and
39:39
um and
39:39
um and what those assets relate to and then
39:42
what those assets relate to and then
39:42
what those assets relate to and then there's kind of the procedural
39:43
there's kind of the procedural
39:43
there's kind of the procedural generation part which is a little bit
39:44
generation part which is a little bit
39:44
generation part which is a little bit more
39:45
more
39:45
more traditional of just sort of number
39:47
traditional of just sort of number
39:48
traditional of just sort of number crunching
39:48
crunching
39:48
crunching okay well i know i know where at the
39:51
okay well i know i know where at the
39:51
okay well i know i know where at the desk is to start with i know where the
39:53
desk is to start with i know where the
39:53
desk is to start with i know where the wall is so
39:53
wall is so
39:53
wall is so therefore i know where to put computers
39:55
therefore i know where to put computers
39:55
therefore i know where to put computers i know where to put i know where to put
39:56
i know where to put i know where to put
39:56
i know where to put i know where to put chairs
39:57
chairs
39:57
chairs and we can just automatically sort of
39:59
and we can just automatically sort of
39:59
and we can just automatically sort of run through that generation
40:00
run through that generation
40:00
run through that generation um as as well and build that up
40:04
um as as well and build that up
40:04
um as as well and build that up in some of the more complex scenarios
40:06
in some of the more complex scenarios
40:06
in some of the more complex scenarios there's also a lot
40:07
there's also a lot
40:07
there's also a lot of intelligence required to actually map
40:10
of intelligence required to actually map
40:10
of intelligence required to actually map textures and especially in the flight
40:12
textures and especially in the flight
40:12
textures and especially in the flight simulator example there's a huge amount
40:14
simulator example there's a huge amount
40:14
simulator example there's a huge amount of
40:15
of
40:15
of knowledge needed to make if you start
40:18
knowledge needed to make if you start
40:18
knowledge needed to make if you start with a basic rectangle which represents
40:20
with a basic rectangle which represents
40:20
with a basic rectangle which represents say
40:22
say
40:22
say a structure you have to know what sort
40:24
a structure you have to know what sort
40:24
a structure you have to know what sort of materials would look
40:26
of materials would look
40:26
of materials would look okay on that whether you know how many
40:28
okay on that whether you know how many
40:28
okay on that whether you know how many windows to use
40:29
windows to use
40:29
windows to use all sorts of different information so
40:31
all sorts of different information so
40:31
all sorts of different information so there's a lot of different techniques
40:33
there's a lot of different techniques
40:33
there's a lot of different techniques um in there depending on your goal i
40:35
um in there depending on your goal i
40:35
um in there depending on your goal i would say
40:37
would say
40:37
would say cool um so the other question that that
40:40
cool um so the other question that that
40:40
cool um so the other question that that people have is
40:41
people have is
40:42
people have is um in in the microsoft example there's a
40:45
um in in the microsoft example there's a
40:45
um in in the microsoft example there's a lot of data that you can use
40:47
lot of data that you can use
40:47
lot of data that you can use um i know for a fact that azure maps has
40:50
um i know for a fact that azure maps has
40:50
um i know for a fact that azure maps has all of these different combinations of
40:53
all of these different combinations of
40:53
all of these different combinations of information for example satellite data
40:55
information for example satellite data
40:55
information for example satellite data geographic data
40:57
geographic data
40:57
geographic data buildings all that stuff but what
41:00
buildings all that stuff but what
41:00
buildings all that stuff but what happens
41:00
happens
41:00
happens if you don't have uh so much data to
41:03
if you don't have uh so much data to
41:03
if you don't have uh so much data to generate
41:05
generate
41:05
generate these extra objects in an indoor space i
41:08
these extra objects in an indoor space i
41:08
these extra objects in an indoor space i can imagine in the case of oven out you
41:09
can imagine in the case of oven out you
41:09
can imagine in the case of oven out you don't have a have a mountain of data but
41:11
don't have a have a mountain of data but
41:11
don't have a have a mountain of data but you have a small pile of data
41:13
you have a small pile of data
41:13
you have a small pile of data to work with well that's that's
41:15
to work with well that's that's
41:15
to work with well that's that's absolutely fair but i mean it's the
41:17
absolutely fair but i mean it's the
41:17
absolutely fair but i mean it's the sort of achilles heel of data and ai is
41:20
sort of achilles heel of data and ai is
41:20
sort of achilles heel of data and ai is we need more data
41:21
we need more data
41:21
we need more data always we need more data so um i think
41:24
always we need more data so um i think
41:24
always we need more data so um i think what microsoft have done behind the
41:25
what microsoft have done behind the
41:25
what microsoft have done behind the scenes
41:26
scenes
41:26
scenes and are continuing to try and do their
41:28
and are continuing to try and do their
41:28
and are continuing to try and do their sort of their work going forward
41:29
sort of their work going forward
41:30
sort of their work going forward is make it as accessible as possible
41:31
is make it as accessible as possible
41:31
is make it as accessible as possible depending on the data type
41:33
depending on the data type
41:33
depending on the data type so we're not dependent on a specific
41:36
so we're not dependent on a specific
41:36
so we're not dependent on a specific format of architectural drawing um you
41:39
format of architectural drawing um you
41:39
format of architectural drawing um you know we're not
41:40
know we're not
41:40
know we're not we're not uh dependent on any specific
41:42
we're not uh dependent on any specific
41:42
we're not uh dependent on any specific satellite data we can just use the data
41:44
satellite data we can just use the data
41:44
satellite data we can just use the data that is available to us
41:45
that is available to us
41:45
that is available to us if at the end of the day um you know
41:48
if at the end of the day um you know
41:48
if at the end of the day um you know when this when the services are released
41:49
when this when the services are released
41:49
when this when the services are released that means
41:50
that means
41:50
that means you're just drawing boxes in an
41:52
you're just drawing boxes in an
41:52
you're just drawing boxes in an architectural plan and saying okay well
41:53
architectural plan and saying okay well
41:53
architectural plan and saying okay well i measured it um you know we can add the
41:56
i measured it um you know we can add the
41:56
i measured it um you know we can add the data that way
41:57
data that way
41:57
data that way but obviously we're in a much stronger
41:59
but obviously we're in a much stronger
41:59
but obviously we're in a much stronger position if we start off
42:00
position if we start off
42:00
position if we start off from a a position with good and reliable
42:03
from a a position with good and reliable
42:03
from a a position with good and reliable data
42:05
data
42:05
data so alicia going back to the conversation
42:08
so alicia going back to the conversation
42:08
so alicia going back to the conversation that we had a few weeks ago
42:09
that we had a few weeks ago
42:09
that we had a few weeks ago you wanted more more windows and there
42:12
you wanted more more windows and there
42:12
you wanted more more windows and there was something with the beach
42:14
was something with the beach
42:14
was something with the beach so well so you know i'm thinking about
42:17
so well so you know i'm thinking about
42:17
so well so you know i'm thinking about it
42:18
it
42:18
it and you know when you look at the
42:20
and you know when you look at the
42:20
and you know when you look at the applications
42:21
applications
42:21
applications if the floor plan is known like is there
42:24
if the floor plan is known like is there
42:24
if the floor plan is known like is there now
42:25
now
42:25
now a future where i can put on a headset
42:28
a future where i can put on a headset
42:28
a future where i can put on a headset and re-skin my work environment so
42:32
and re-skin my work environment so
42:32
and re-skin my work environment so so you you can put me in the warehouse
42:34
so you you can put me in the warehouse
42:34
so you you can put me in the warehouse or you can put me in the basement
42:36
or you can put me in the basement
42:36
or you can put me in the basement and you know i can go ahead and paint
42:39
and you know i can go ahead and paint
42:39
and you know i can go ahead and paint the walls bright pink if i want
42:41
the walls bright pink if i want
42:41
the walls bright pink if i want and my cubicle partner can paint the
42:44
and my cubicle partner can paint the
42:44
and my cubicle partner can paint the walls
42:45
walls
42:45
walls you know black and gray if they want
42:48
you know black and gray if they want
42:48
you know black and gray if they want and uh maybe there's some re-skinning
42:51
and uh maybe there's some re-skinning
42:52
and uh maybe there's some re-skinning that we can do
42:53
that we can do
42:53
that we can do for our environment yeah that's a great
42:56
for our environment yeah that's a great
42:56
for our environment yeah that's a great great idea yeah then you know if they
42:59
great idea yeah then you know if they
42:59
great idea yeah then you know if they could just tie that to the sensation
43:01
could just tie that to the sensation
43:01
could just tie that to the sensation of you know jumping into the water
43:04
of you know jumping into the water
43:04
of you know jumping into the water then i would be set right yeah
43:09
then i would be set right yeah
43:09
then i would be set right yeah yeah i mean if we could bring this down
43:11
yeah i mean if we could bring this down
43:11
yeah i mean if we could bring this down from architectural drawings and a lot of
43:13
from architectural drawings and a lot of
43:13
from architectural drawings and a lot of complicated technical stuff
43:15
complicated technical stuff
43:15
complicated technical stuff down to a png i can draw a png in paint
43:19
down to a png i can draw a png in paint
43:19
down to a png i can draw a png in paint i mean just draw a few lines and we're
43:21
i mean just draw a few lines and we're
43:21
i mean just draw a few lines and we're good we're set
43:22
good we're set
43:22
good we're set we can design an office i guess or or
43:25
we can design an office i guess or or
43:25
we can design an office i guess or or does it take more than that actually
43:29
well on the on the sort of base level no
43:32
well on the on the sort of base level no
43:32
well on the on the sort of base level no if you have that basic
43:34
if you have that basic
43:34
if you have that basic basic drawing we can give you we can
43:35
basic drawing we can give you we can
43:35
basic drawing we can give you we can give you a basic building and once
43:37
give you a basic building and once
43:37
give you a basic building and once you're building
43:38
you're building
43:38
you're building is or wants your virtual environment is
43:40
is or wants your virtual environment is
43:40
is or wants your virtual environment is set up then depending on the exact tools
43:42
set up then depending on the exact tools
43:42
set up then depending on the exact tools you're using
43:43
you're using
43:43
you're using you can just go in either to unity so in
43:45
you can just go in either to unity so in
43:45
you can just go in either to unity so in the unity example you could just drop
43:47
the unity example you could just drop
43:47
the unity example you could just drop drag and drop colors onto individual
43:49
drag and drop colors onto individual
43:49
drag and drop colors onto individual walls
43:49
walls
43:49
walls so for example there's a behind me this
43:52
so for example there's a behind me this
43:52
so for example there's a behind me this way there's a yellow wall
43:53
way there's a yellow wall
43:53
way there's a yellow wall um if we wanted to change that pink i
43:55
um if we wanted to change that pink i
43:55
um if we wanted to change that pink i could just go into unity and drag a pink
43:56
could just go into unity and drag a pink
43:56
could just go into unity and drag a pink color and personalize that so
43:58
color and personalize that so
43:58
color and personalize that so they can be personalized um again if
44:00
they can be personalized um again if
44:00
they can be personalized um again if it's a shared virtual reality space then
44:02
it's a shared virtual reality space then
44:02
it's a shared virtual reality space then we probably have to think about the
44:03
we probably have to think about the
44:03
we probably have to think about the other people but it's our own
44:04
other people but it's our own
44:04
other people but it's our own version of the office and that's the
44:06
version of the office and that's the
44:06
version of the office and that's the great thing about the
44:08
great thing about the
44:08
great thing about the sort of personalization part is we can
44:10
sort of personalization part is we can
44:10
sort of personalization part is we can do whatever we want
44:11
do whatever we want
44:11
do whatever we want the other thing is we can actually uh
44:13
the other thing is we can actually uh
44:13
the other thing is we can actually uh you know using
44:14
you know using
44:14
you know using the headset we can drag in tools
44:17
the headset we can drag in tools
44:17
the headset we can drag in tools um collaboration tools we can drag in
44:20
um collaboration tools we can drag in
44:20
um collaboration tools we can drag in video players we can drag in
44:22
video players we can drag in
44:22
video players we can drag in games to play together in vr as well as
44:25
games to play together in vr as well as
44:25
games to play together in vr as well as like customizing our environment in the
44:26
like customizing our environment in the
44:26
like customizing our environment in the way it looks we can actually customize
44:28
way it looks we can actually customize
44:28
way it looks we can actually customize it in the way that we can interact with
44:29
it in the way that we can interact with
44:29
it in the way that we can interact with it which is really really
44:31
it which is really really
44:32
it which is really really so today we have joining us anthony
44:35
so today we have joining us anthony
44:35
so today we have joining us anthony bartolo from microsoft
44:36
bartolo from microsoft
44:36
bartolo from microsoft and uh anthony how are you
44:40
and uh anthony how are you
44:40
and uh anthony how are you thank you for joining us
44:44
and we're gonna have him check his sound
44:47
and we're gonna have him check his sound
44:47
and we're gonna have him check his sound really quickly um
44:50
really quickly um
44:50
really quickly um but but anthony does a lot of great work
44:53
but but anthony does a lot of great work
44:53
but but anthony does a lot of great work integrating
44:54
integrating
44:54
integrating uh drones and hololens and
44:58
uh drones and hololens and
44:58
uh drones and hololens and solves some similar problems for users
45:01
solves some similar problems for users
45:01
solves some similar problems for users so i
45:02
so i
45:02
so i i i'd love to hear anthony's
45:05
i i'd love to hear anthony's
45:05
i i'd love to hear anthony's input and you know how he the
45:08
input and you know how he the
45:08
input and you know how he the applications for this technology and
45:10
applications for this technology and
45:10
applications for this technology and some of these
45:11
some of these
45:11
some of these problems that are being solved well
45:13
problems that are being solved well
45:13
problems that are being solved well actually
45:14
actually
45:14
actually i'd like to discuss a little bit further
45:15
i'd like to discuss a little bit further
45:15
i'd like to discuss a little bit further in terms of this solution because
45:17
in terms of this solution because
45:17
in terms of this solution because what i love about it is it addresses an
45:19
what i love about it is it addresses an
45:19
what i love about it is it addresses an opportunity you know too many times
45:21
opportunity you know too many times
45:21
opportunity you know too many times we've seen organizations go forward
45:23
we've seen organizations go forward
45:23
we've seen organizations go forward and say how do i adopt this new
45:25
and say how do i adopt this new
45:25
and say how do i adopt this new technology and
45:26
technology and
45:26
technology and add it to something instead of how do i
45:29
add it to something instead of how do i
45:29
add it to something instead of how do i address
45:30
address
45:30
address a problem or an opportunity uh
45:32
a problem or an opportunity uh
45:32
a problem or an opportunity uh leveraging technology but not having
45:34
leveraging technology but not having
45:34
leveraging technology but not having technology being the main purpose as to
45:36
technology being the main purpose as to
45:36
technology being the main purpose as to why the adoption is occurring
45:38
why the adoption is occurring
45:38
why the adoption is occurring what i loved about the demonstration
45:39
what i loved about the demonstration
45:39
what i loved about the demonstration that was just delivered was the proper
45:42
that was just delivered was the proper
45:42
that was just delivered was the proper use of you know office space in terms of
45:46
use of you know office space in terms of
45:46
use of you know office space in terms of interacting with employees there's you
45:48
interacting with employees there's you
45:48
interacting with employees there's you know there's a need for
45:49
know there's a need for
45:49
know there's a need for increasing of engagement there is a need
45:52
increasing of engagement there is a need
45:52
increasing of engagement there is a need for you know increase an
45:53
for you know increase an
45:54
for you know increase an increase of playfulness amongst
45:57
increase of playfulness amongst
45:57
increase of playfulness amongst coworkers and so that instance and that
46:00
coworkers and so that instance and that
46:00
coworkers and so that instance and that capability that was highlighted
46:02
capability that was highlighted
46:02
capability that was highlighted was done in such an awesome way in the
46:03
was done in such an awesome way in the
46:03
was done in such an awesome way in the fact that it highlights how
46:05
fact that it highlights how
46:05
fact that it highlights how technology uplifts um the the
46:08
technology uplifts um the the
46:08
technology uplifts um the the opportunity or problem that needs to be
46:09
opportunity or problem that needs to be
46:09
opportunity or problem that needs to be addressed
46:10
addressed
46:10
addressed as opposed to being the main cornerstone
46:12
as opposed to being the main cornerstone
46:12
as opposed to being the main cornerstone behind it
46:14
behind it
46:14
behind it so i know you've been working on a
46:17
so i know you've been working on a
46:17
so i know you've been working on a project to
46:18
project to
46:18
project to in the search and rescue area
46:21
in the search and rescue area
46:21
in the search and rescue area i can imagine that it would be useful if
46:24
i can imagine that it would be useful if
46:24
i can imagine that it would be useful if you wanted to
46:25
you wanted to
46:25
you wanted to build a search and rescue drone to have
46:28
build a search and rescue drone to have
46:28
build a search and rescue drone to have a sort of a 3d image of the environment
46:30
a sort of a 3d image of the environment
46:30
a sort of a 3d image of the environment that the drone is flying in
46:33
that the drone is flying in
46:33
that the drone is flying in is it something that you do you actually
46:36
is it something that you do you actually
46:36
is it something that you do you actually tried in your solution i mean people
46:39
tried in your solution i mean people
46:39
tried in your solution i mean people will get to introduce
46:40
will get to introduce
46:40
will get to introduce the solution itself later on so this
46:41
the solution itself later on so this
46:41
the solution itself later on so this might be a bit unclear to them but
46:43
might be a bit unclear to them but
46:43
might be a bit unclear to them but um i'm curious uh how can you actually
46:46
um i'm curious uh how can you actually
46:46
um i'm curious uh how can you actually use
46:47
use
46:47
use 2d maps for controlling drones in this
46:51
2d maps for controlling drones in this
46:51
2d maps for controlling drones in this case
46:52
case
46:52
case and that's an interesting premise right
46:53
and that's an interesting premise right
46:53
and that's an interesting premise right behind what what can be accomplished
46:56
behind what what can be accomplished
46:56
behind what what can be accomplished back when this proof of concept was
46:57
back when this proof of concept was
46:57
back when this proof of concept was actually initialized
46:59
actually initialized
46:59
actually initialized this whole aspect of spatial mapping
47:01
this whole aspect of spatial mapping
47:01
this whole aspect of spatial mapping wasn't available the technology didn't
47:03
wasn't available the technology didn't
47:03
wasn't available the technology didn't exist
47:03
exist
47:03
exist uh in respect to understanding the
47:05
uh in respect to understanding the
47:06
uh in respect to understanding the surrounding areas
47:07
surrounding areas
47:07
surrounding areas the drone itself was taught to pinpoint
47:10
the drone itself was taught to pinpoint
47:10
the drone itself was taught to pinpoint a specific instance or
47:11
a specific instance or
47:11
a specific instance or issue which is in essence the life
47:13
issue which is in essence the life
47:13
issue which is in essence the life jacket uh inside of the uh
47:14
jacket uh inside of the uh
47:14
jacket uh inside of the uh on the water to fly to that location
47:18
on the water to fly to that location
47:18
on the water to fly to that location the requirement was such that it was you
47:20
the requirement was such that it was you
47:20
the requirement was such that it was you know giving the coordinates to the drone
47:22
know giving the coordinates to the drone
47:22
know giving the coordinates to the drone the drone would specifically apply to
47:24
the drone would specifically apply to
47:24
the drone would specifically apply to that geographical
47:25
that geographical
47:25
that geographical coordinate uh the gps coordinates that
47:27
coordinate uh the gps coordinates that
47:27
coordinate uh the gps coordinates that was given to it and then
47:29
was given to it and then
47:29
was given to it and then once there assessed the situation or the
47:31
once there assessed the situation or the
47:31
once there assessed the situation or the area right
47:32
area right
47:32
area right even to that respect you know what was
47:35
even to that respect you know what was
47:35
even to that respect you know what was done to make that
47:36
done to make that
47:36
done to make that happen was based on the opportunity not
47:39
happen was based on the opportunity not
47:39
happen was based on the opportunity not on the technology
47:40
on the technology
47:40
on the technology the technology in terms of custom vision
47:42
the technology in terms of custom vision
47:42
the technology in terms of custom vision for in as an example
47:43
for in as an example
47:43
for in as an example didn't even exist in its current form uh
47:46
didn't even exist in its current form uh
47:46
didn't even exist in its current form uh and so when we built out this model was
47:49
and so when we built out this model was
47:49
and so when we built out this model was it had to be on need of
47:51
it had to be on need of
47:51
it had to be on need of the solution that we were trying to
47:52
the solution that we were trying to
47:52
the solution that we were trying to address not how do we incorporate this
47:54
address not how do we incorporate this
47:54
address not how do we incorporate this to
47:55
to
47:55
to to make this better right so it was it
47:57
to make this better right so it was it
47:57
to make this better right so it was it was an interesting premise
47:58
was an interesting premise
47:58
was an interesting premise in terms of how we understand how
48:01
in terms of how we understand how
48:01
in terms of how we understand how technology is leveraged to
48:02
technology is leveraged to
48:02
technology is leveraged to address opportunities and problems as
48:05
address opportunities and problems as
48:05
address opportunities and problems as opposed to adding it for the sake of
48:06
opposed to adding it for the sake of
48:06
opposed to adding it for the sake of valuable
48:07
valuable
48:07
valuable but definitely you know in the spatial
48:08
but definitely you know in the spatial
48:08
but definitely you know in the spatial mapping sense uh had the technology been
48:11
mapping sense uh had the technology been
48:11
mapping sense uh had the technology been available for the proven concept
48:12
available for the proven concept
48:12
available for the proven concept 100 would have been it would have been a
48:14
100 would have been it would have been a
48:14
100 would have been it would have been a huge advantage uh
48:16
huge advantage uh
48:16
huge advantage uh more so from from a further automation
48:18
more so from from a further automation
48:18
more so from from a further automation of the drone flying itself
48:20
of the drone flying itself
48:20
of the drone flying itself yeah it sounds really interesting so
48:22
yeah it sounds really interesting so
48:22
yeah it sounds really interesting so fergus
48:23
fergus
48:23
fergus i was wondering um how did you
48:27
i was wondering um how did you
48:27
i was wondering um how did you actually start this project did you
48:28
actually start this project did you
48:28
actually start this project did you start from a
48:30
start from a
48:30
start from a technology perspective uh
48:34
technology perspective uh
48:34
technology perspective uh in in a sense that you wanted to use
48:37
in in a sense that you wanted to use
48:37
in in a sense that you wanted to use uh emerging technologies and then you
48:40
uh emerging technologies and then you
48:40
uh emerging technologies and then you found a problem or did you start from
48:41
found a problem or did you start from
48:42
found a problem or did you start from the problem i'm really interested in
48:43
the problem i'm really interested in
48:44
the problem i'm really interested in in how does a project like this come to
48:45
in how does a project like this come to
48:46
in how does a project like this come to be
48:47
be
48:47
be yeah that's a really uh great question
48:48
yeah that's a really uh great question
48:48
yeah that's a really uh great question actually so the project came to be
48:51
actually so the project came to be
48:51
actually so the project came to be because we were building a new space and
48:53
because we were building a new space and
48:53
because we were building a new space and we had an opportunity to
48:55
we had an opportunity to
48:55
we had an opportunity to start from scratch and accenture thought
48:57
start from scratch and accenture thought
48:57
start from scratch and accenture thought okay let's do
48:58
okay let's do
48:58
okay let's do this right in the right way let's
49:02
this right in the right way let's
49:02
this right in the right way let's do you know build a building that
49:04
do you know build a building that
49:04
do you know build a building that represents our technological
49:05
represents our technological
49:05
represents our technological capabilities
49:07
capabilities
49:07
capabilities um our partnership with microsoft as
49:09
um our partnership with microsoft as
49:09
um our partnership with microsoft as well as other vendors and suppliers
49:11
well as other vendors and suppliers
49:11
well as other vendors and suppliers and really shows off the future of work
49:14
and really shows off the future of work
49:14
and really shows off the future of work so that's kind of how the
49:16
so that's kind of how the
49:16
so that's kind of how the the project came about it was more
49:18
the project came about it was more
49:18
the project came about it was more putting people first and technology and
49:20
putting people first and technology and
49:20
putting people first and technology and making sure that these spaces were
49:22
making sure that these spaces were
49:22
making sure that these spaces were accessible usable
49:24
accessible usable
49:24
accessible usable and you know using technology to
49:29
and you know using technology to
49:29
and you know using technology to make that happen so really we went about
49:31
make that happen so really we went about
49:31
make that happen so really we went about choosing the
49:33
choosing the
49:33
choosing the the best technologies to fit that
49:34
the best technologies to fit that
49:34
the best technologies to fit that particular need obviously
49:36
particular need obviously
49:36
particular need obviously a few months into that the pandemic hits
49:38
a few months into that the pandemic hits
49:38
a few months into that the pandemic hits and things start to change a little bit
49:40
and things start to change a little bit
49:40
and things start to change a little bit and then we start
49:41
and then we start
49:41
and then we start asking ourselves oh actually the
49:43
asking ourselves oh actually the
49:43
asking ourselves oh actually the technologies that we're using in the
49:44
technologies that we're using in the
49:44
technologies that we're using in the solution that we're building can
49:46
solution that we're building can
49:46
solution that we're building can actually be used
49:47
actually be used
49:47
actually be used further so it sort of evolved as we went
49:49
further so it sort of evolved as we went
49:49
further so it sort of evolved as we went on and that's the
49:50
on and that's the
49:50
on and that's the great part of working in a research team
49:53
great part of working in a research team
49:53
great part of working in a research team um
49:54
um
49:54
um with is you're slightly flexible to
49:55
with is you're slightly flexible to
49:55
with is you're slightly flexible to explore those new ideas and build new
49:57
explore those new ideas and build new
49:57
explore those new ideas and build new assets
49:58
assets
49:58
assets and do some of that development but
49:59
and do some of that development but
49:59
and do some of that development but really it came from a sort of a people
50:01
really it came from a sort of a people
50:01
really it came from a sort of a people opportunity first
50:03
opportunity first
50:04
opportunity first yeah nice
50:07
yeah nice
50:07
yeah nice um alicia i
50:11
um alicia i
50:11
um alicia i i guess that um so we've we've talked
50:13
i guess that um so we've we've talked
50:13
i guess that um so we've we've talked about
50:14
about
50:14
about our beach thing um uh
50:17
our beach thing um uh
50:17
our beach thing um uh uh i was thinking maybe um
50:21
uh i was thinking maybe um
50:21
uh i was thinking maybe um we could do a virtual reality fly by um
50:25
we could do a virtual reality fly by um
50:25
we could do a virtual reality fly by um in a drone or something and maybe
50:27
in a drone or something and maybe
50:27
in a drone or something and maybe control it from our vr
50:28
control it from our vr
50:28
control it from our vr set is that is that a far-fetched idea
50:31
set is that is that a far-fetched idea
50:31
set is that is that a far-fetched idea once we have this
50:32
once we have this
50:32
once we have this this combination of technologies or uh
50:35
this combination of technologies or uh
50:35
this combination of technologies or uh how does that work
50:35
how does that work
50:36
how does that work anthony oh
50:39
anthony oh
50:39
anthony oh anthony or alicia i thought you were
50:40
anthony or alicia i thought you were
50:40
anthony or alicia i thought you were talking to him
50:43
all right sorry my head is in the clouds
50:45
all right sorry my head is in the clouds
50:45
all right sorry my head is in the clouds i just booked my tickets to hawaii
50:47
i just booked my tickets to hawaii
50:47
i just booked my tickets to hawaii this week so no serious talk for me
50:53
this week so no serious talk for me
50:53
this week so no serious talk for me so my background is actually maltese and
50:55
so my background is actually maltese and
50:55
so my background is actually maltese and i haven't been back to malta since 2000
50:57
i haven't been back to malta since 2000
50:57
i haven't been back to malta since 2000 and one thing i was thinking about in
51:00
and one thing i was thinking about in
51:00
and one thing i was thinking about in terms of the whole project that we did
51:01
terms of the whole project that we did
51:01
terms of the whole project that we did with the drone
51:02
with the drone
51:02
with the drone was in the inclusion of spatial mapping
51:04
was in the inclusion of spatial mapping
51:04
was in the inclusion of spatial mapping and tools like you know
51:06
and tools like you know
51:06
and tools like you know the flight simulator um piece in in
51:08
the flight simulator um piece in in
51:08
the flight simulator um piece in in terms of the mapping of the
51:09
terms of the mapping of the
51:09
terms of the mapping of the the geos and the understanding of your
51:11
the geos and the understanding of your
51:11
the geos and the understanding of your space how far
51:13
space how far
51:13
space how far are we from walking on the beach
51:15
are we from walking on the beach
51:15
are we from walking on the beach virtually via our headsets
51:17
virtually via our headsets
51:17
virtually via our headsets um and having in our place on ramla bay
51:20
um and having in our place on ramla bay
51:20
um and having in our place on ramla bay as an example in
51:22
as an example in
51:22
as an example in in uh bengozo uh the ability to walk
51:25
in uh bengozo uh the ability to walk
51:25
in uh bengozo uh the ability to walk down the beach
51:26
down the beach
51:26
down the beach aside from touch you know even smell
51:29
aside from touch you know even smell
51:29
aside from touch you know even smell right now can be
51:30
right now can be
51:30
right now can be simulated in terms of you know the
51:32
simulated in terms of you know the
51:32
simulated in terms of you know the devices that we had and smelling
51:33
devices that we had and smelling
51:34
devices that we had and smelling of the salt water and the sand and what
51:35
of the salt water and the sand and what
51:35
of the salt water and the sand and what have you right so
51:37
have you right so
51:37
have you right so you know we're not that far away uh the
51:40
you know we're not that far away uh the
51:40
you know we're not that far away uh the the technology exists
51:41
the technology exists
51:41
the technology exists it's just the problem itself or the
51:43
it's just the problem itself or the
51:43
it's just the problem itself or the opportunity has to be solved
51:45
opportunity has to be solved
51:45
opportunity has to be solved right so so you know
51:48
right so so you know
51:48
right so so you know it all is based on our imagination our
51:51
it all is based on our imagination our
51:51
it all is based on our imagination our creativity
51:52
creativity
51:52
creativity but doing so in such a way that we're
51:53
but doing so in such a way that we're
51:53
but doing so in such a way that we're addressing that problem or opportunity
51:55
addressing that problem or opportunity
51:55
addressing that problem or opportunity not just rubbing technology on it in the
51:57
not just rubbing technology on it in the
51:57
not just rubbing technology on it in the hope that
51:57
hope that
51:57
hope that it's gonna make something really cool
52:01
yeah yeah so um
52:04
yeah yeah so um
52:04
yeah yeah so um especially given the whole kovitz
52:06
especially given the whole kovitz
52:06
especially given the whole kovitz situation
52:07
situation
52:07
situation i'm thinking that this could be very
52:09
i'm thinking that this could be very
52:09
i'm thinking that this could be very very useful
52:10
very useful
52:10
very useful um um so alicia and i have been joking
52:13
um um so alicia and i have been joking
52:14
um um so alicia and i have been joking around
52:15
around
52:15
around about this stuff for for days actually
52:17
about this stuff for for days actually
52:17
about this stuff for for days actually uh
52:18
uh
52:18
uh every every time i speak to her on on
52:20
every every time i speak to her on on
52:20
every every time i speak to her on on slack on on
52:21
slack on on
52:21
slack on on teams or other online tools we keep
52:24
teams or other online tools we keep
52:24
teams or other online tools we keep getting back to this idea having virtual
52:27
getting back to this idea having virtual
52:27
getting back to this idea having virtual reality and ai combined
52:30
reality and ai combined
52:30
reality and ai combined and generating a space where you can get
52:32
and generating a space where you can get
52:32
and generating a space where you can get outside during a lockdown that sounds
52:34
outside during a lockdown that sounds
52:34
outside during a lockdown that sounds really
52:35
really
52:35
really for some people it might actually be
52:37
for some people it might actually be
52:37
for some people it might actually be life-saving i know from from
52:40
life-saving i know from from
52:40
life-saving i know from from a couple of friends that um they can be
52:43
a couple of friends that um they can be
52:43
a couple of friends that um they can be they can feel
52:43
they can feel
52:43
they can feel pretty locked up these days and
52:46
pretty locked up these days and
52:46
pretty locked up these days and and it can have severe psychological
52:50
and it can have severe psychological
52:50
and it can have severe psychological uh impact on them so i'm thinking yeah
52:52
uh impact on them so i'm thinking yeah
52:52
uh impact on them so i'm thinking yeah this this could be
52:53
this this could be
52:53
this this could be actually a tool not only for just for
52:56
actually a tool not only for just for
52:56
actually a tool not only for just for fun
52:57
fun
52:57
fun like alicia and i have been joking about
52:59
like alicia and i have been joking about
53:00
like alicia and i have been joking about it
53:00
it
53:00
it but actually for real and that makes it
53:02
but actually for real and that makes it
53:02
but actually for real and that makes it really interesting
53:03
really interesting
53:03
really interesting especially in this day and age so
53:06
especially in this day and age so
53:06
especially in this day and age so what and this is a question for both of
53:08
what and this is a question for both of
53:08
what and this is a question for both of you anthony and
53:09
you anthony and
53:09
you anthony and fergus what do you think is the most
53:12
fergus what do you think is the most
53:12
fergus what do you think is the most important piece of technology that
53:14
important piece of technology that
53:14
important piece of technology that you got that made you decide that this
53:17
you got that made you decide that this
53:17
you got that made you decide that this project is feasible
53:21
anthony any any thoughts so
53:24
anthony any any thoughts so
53:24
anthony any any thoughts so this is the thing right when we did the
53:27
this is the thing right when we did the
53:27
this is the thing right when we did the drone project
53:27
drone project
53:28
drone project for for on behalf of the king uh coast
53:30
for for on behalf of the king uh coast
53:30
for for on behalf of the king uh coast guard with
53:31
guard with
53:31
guard with uh partner indro robotics out of
53:33
uh partner indro robotics out of
53:33
uh partner indro robotics out of vancouver
53:34
vancouver
53:34
vancouver the technology didn't exist right what
53:37
the technology didn't exist right what
53:37
the technology didn't exist right what we were asked to do
53:38
we were asked to do
53:38
we were asked to do was the the drones that were flying out
53:40
was the the drones that were flying out
53:40
was the the drones that were flying out to sea were were men
53:42
to sea were were men
53:42
to sea were were men man controlled uh or operator controlled
53:44
man controlled uh or operator controlled
53:44
man controlled uh or operator controlled sorry
53:45
sorry
53:45
sorry and you had this drone that would go out
53:47
and you had this drone that would go out
53:47
and you had this drone that would go out and survey the area
53:49
and survey the area
53:49
and survey the area and then film and then fly back the
53:51
and then film and then fly back the
53:51
and then film and then fly back the initial ass that had come to microsoft
53:53
initial ass that had come to microsoft
53:53
initial ass that had come to microsoft was
53:54
was
53:54
was we need you to help us with the
53:56
we need you to help us with the
53:56
we need you to help us with the compression agent
53:57
compression agent
53:57
compression agent uh to compress the video stream because
53:59
uh to compress the video stream because
53:59
uh to compress the video stream because right now there is no connectivity for
54:01
right now there is no connectivity for
54:01
right now there is no connectivity for this drone that's flying out there
54:03
this drone that's flying out there
54:03
this drone that's flying out there and we have to wait till the drone comes
54:04
and we have to wait till the drone comes
54:04
and we have to wait till the drone comes back into the central office to review
54:05
back into the central office to review
54:05
back into the central office to review the tape
54:06
the tape
54:06
the tape uh and to do the analytics on what the
54:08
uh and to do the analytics on what the
54:08
uh and to do the analytics on what the situation is so that we can send out
54:10
situation is so that we can send out
54:10
situation is so that we can send out rescue
54:10
rescue
54:10
rescue uh equipment out there to save these
54:12
uh equipment out there to save these
54:12
uh equipment out there to save these people and
54:13
people and
54:13
people and in you know saving a life every every
54:15
in you know saving a life every every
54:15
in you know saving a life every every second counts and so
54:17
second counts and so
54:17
second counts and so you know when we looked at this and we
54:18
you know when we looked at this and we
54:18
you know when we looked at this and we looked at the opportunity like i said
54:20
looked at the opportunity like i said
54:20
looked at the opportunity like i said computer vision was was wasn't even
54:22
computer vision was was wasn't even
54:22
computer vision was was wasn't even launched yet it was still in its infancy
54:24
launched yet it was still in its infancy
54:24
launched yet it was still in its infancy in terms of what microsoft was testing
54:25
in terms of what microsoft was testing
54:25
in terms of what microsoft was testing with
54:26
with
54:26
with and we you know thankfully made a
54:28
and we you know thankfully made a
54:28
and we you know thankfully made a connection with the engineers here at
54:29
connection with the engineers here at
54:29
connection with the engineers here at microsoft
54:30
microsoft
54:30
microsoft uh to to allow us to run this test and
54:32
uh to to allow us to run this test and
54:32
uh to to allow us to run this test and run this functionality to address this
54:34
run this functionality to address this
54:34
run this functionality to address this opportunity
54:34
opportunity
54:34
opportunity to see if it was even feasible what's
54:37
to see if it was even feasible what's
54:37
to see if it was even feasible what's been amazing
54:38
been amazing
54:38
been amazing is the amount of growth has occurred
54:41
is the amount of growth has occurred
54:41
is the amount of growth has occurred from custom vision
54:43
from custom vision
54:43
from custom vision from the perspective of the initial
54:44
from the perspective of the initial
54:44
from the perspective of the initial thought was you know just recognizing
54:46
thought was you know just recognizing
54:46
thought was you know just recognizing objects
54:46
objects
54:46
objects there was no thought of a moving object
54:49
there was no thought of a moving object
54:49
there was no thought of a moving object recognizing objects or
54:50
recognizing objects or
54:50
recognizing objects or you know let alone understanding what
54:53
you know let alone understanding what
54:53
you know let alone understanding what objects look like on the edge without
54:55
objects look like on the edge without
54:55
objects look like on the edge without connectivity
54:56
connectivity
54:56
connectivity right we talk about uh edge computing
54:58
right we talk about uh edge computing
54:58
right we talk about uh edge computing today
54:59
today
54:59
today the the proof of concept that we started
55:00
the the proof of concept that we started
55:00
the the proof of concept that we started out with with the drone project
55:02
out with with the drone project
55:02
out with with the drone project was four years ago um so so
55:05
was four years ago um so so
55:05
was four years ago um so so you know the way the technology moves so
55:08
you know the way the technology moves so
55:08
you know the way the technology moves so fast and so quickly
55:09
fast and so quickly
55:09
fast and so quickly i love the fact that technology is is
55:11
i love the fact that technology is is
55:11
i love the fact that technology is is adapting to what our needs are
55:14
adapting to what our needs are
55:14
adapting to what our needs are as opposed to organizations dictating
55:16
as opposed to organizations dictating
55:16
as opposed to organizations dictating where technology
55:17
where technology
55:17
where technology is to go yeah
55:21
is to go yeah
55:21
is to go yeah and that i would absolutely echo that as
55:23
and that i would absolutely echo that as
55:23
and that i would absolutely echo that as well i think um
55:24
well i think um
55:24
well i think um you know for us it's really a
55:26
you know for us it's really a
55:26
you know for us it's really a combination of technologies
55:28
combination of technologies
55:28
combination of technologies some the iot devices the cloud
55:30
some the iot devices the cloud
55:30
some the iot devices the cloud infrastructure has been there for a
55:31
infrastructure has been there for a
55:31
infrastructure has been there for a while but
55:32
while but
55:32
while but the whole combination of integrating
55:34
the whole combination of integrating
55:34
the whole combination of integrating that with a familiar
55:36
that with a familiar
55:36
that with a familiar visualization platform in the 2d maps is
55:38
visualization platform in the 2d maps is
55:38
visualization platform in the 2d maps is really only over the last year or so
55:41
really only over the last year or so
55:41
really only over the last year or so and it's just a sort of combination of
55:42
and it's just a sort of combination of
55:42
and it's just a sort of combination of all the different technologies
55:44
all the different technologies
55:44
all the different technologies as well as the the technologies
55:46
as well as the the technologies
55:46
as well as the the technologies themselves being more available
55:48
themselves being more available
55:48
themselves being more available so vr headsets are coming down in price
55:53
so vr headsets are coming down in price
55:53
so vr headsets are coming down in price they're getting newer versions which are
55:55
they're getting newer versions which are
55:55
they're getting newer versions which are lighter
55:56
lighter
55:56
lighter you know you can use your mobile phone
55:57
you know you can use your mobile phone
55:57
you know you can use your mobile phone in a cardboard frame
55:59
in a cardboard frame
55:59
in a cardboard frame to access some of these these
56:02
to access some of these these
56:02
to access some of these these functionalities and these features
56:04
functionalities and these features
56:04
functionalities and these features uh so the you know the entry cost is
56:06
uh so the you know the entry cost is
56:06
uh so the you know the entry cost is coming down which makes it more
56:08
coming down which makes it more
56:08
coming down which makes it more accessible
56:09
accessible
56:09
accessible in terms of your question around like
56:12
in terms of your question around like
56:12
in terms of your question around like how you know engaging it is
56:16
how you know engaging it is
56:16
how you know engaging it is and things i think one thing we were
56:18
and things i think one thing we were
56:18
and things i think one thing we were looking at which is quite interesting
56:19
looking at which is quite interesting
56:19
looking at which is quite interesting during our testing
56:21
during our testing
56:21
during our testing was the sense of physical proximity and
56:22
was the sense of physical proximity and
56:22
was the sense of physical proximity and how important that was to actually
56:24
how important that was to actually
56:24
how important that was to actually feeling like you're using space
56:25
feeling like you're using space
56:25
feeling like you're using space together and using um
56:29
together and using um
56:29
together and using um objects and items in the physical space
56:30
objects and items in the physical space
56:30
objects and items in the physical space together so i think we're all quite
56:33
together so i think we're all quite
56:33
together so i think we're all quite familiar with
56:33
familiar with
56:33
familiar with you know teams calls and zoom calls uh
56:36
you know teams calls and zoom calls uh
56:36
you know teams calls and zoom calls uh and
56:37
and
56:37
and for our socializing these days since
56:39
for our socializing these days since
56:39
for our socializing these days since it's
56:40
it's
56:40
it's more difficult to get out of the house
56:42
more difficult to get out of the house
56:42
more difficult to get out of the house but the
56:44
but the
56:44
but the the addition of extended reality allows
56:46
the addition of extended reality allows
56:46
the addition of extended reality allows you to actually get that physical
56:47
you to actually get that physical
56:48
you to actually get that physical sense of presence so if i sort of move
56:50
sense of presence so if i sort of move
56:50
sense of presence so if i sort of move my hand
56:51
my hand
56:51
my hand sharply at the camera very few of you at
56:53
sharply at the camera very few of you at
56:53
sharply at the camera very few of you at home are going to react but if we're
56:54
home are going to react but if we're
56:54
home are going to react but if we're together in extended reality and i move
56:56
together in extended reality and i move
56:56
together in extended reality and i move my hand very sharply at your face you're
56:58
my hand very sharply at your face you're
56:58
my hand very sharply at your face you're probably going to recoil
56:59
probably going to recoil
56:59
probably going to recoil and move back because your brain
57:00
and move back because your brain
57:00
and move back because your brain actually sort of processes the
57:01
actually sort of processes the
57:02
actually sort of processes the information differently
57:04
information differently
57:04
information differently yeah so so alicia
57:08
yeah so so alicia
57:08
yeah so so alicia this is an interesting question for you
57:10
this is an interesting question for you
57:10
this is an interesting question for you i know you've got an oculus uh
57:12
i know you've got an oculus uh
57:12
i know you've got an oculus uh quest uh at home um have you tried any
57:15
quest uh at home um have you tried any
57:15
quest uh at home um have you tried any of the other devices so
57:17
of the other devices so
57:17
of the other devices so like the one uh we just mentioned the
57:19
like the one uh we just mentioned the
57:19
like the one uh we just mentioned the cardboard box in your mobile phone lined
57:21
cardboard box in your mobile phone lined
57:21
cardboard box in your mobile phone lined up
57:22
up
57:22
up um i i haven't yet so um
57:26
um i i haven't yet so um
57:26
um i i haven't yet so um what what sort of experience do you get
57:28
what what sort of experience do you get
57:28
what what sort of experience do you get from from the
57:29
from from the
57:29
from from the device that you have currently the uh
57:31
device that you have currently the uh
57:31
device that you have currently the uh the oculus quest how how's that
57:33
the oculus quest how how's that
57:33
the oculus quest how how's that uh for you um it's so
57:36
uh for you um it's so
57:36
uh for you um it's so uh i we this is actually a
57:40
uh i we this is actually a
57:40
uh i we this is actually a giveaway item from one of our last
57:42
giveaway item from one of our last
57:42
giveaway item from one of our last sessions um
57:43
sessions um
57:43
sessions um so i haven't opened the box i've been
57:45
so i haven't opened the box i've been
57:45
so i haven't opened the box i've been really good i've been dying
57:47
really good i've been dying
57:47
really good i've been dying to to open the box it's a an open
57:50
to to open the box it's a an open
57:50
to to open the box it's a an open present that's sitting in my office
57:52
present that's sitting in my office
57:52
present that's sitting in my office right now
57:53
right now
57:53
right now um but um anthony and
57:56
um but um anthony and
57:56
um but um anthony and fergus if you had uh any feedback on
57:59
fergus if you had uh any feedback on
57:59
fergus if you had uh any feedback on devices out there
58:02
i i've got a barrage of devices behind
58:04
i i've got a barrage of devices behind
58:04
i i've got a barrage of devices behind me i don't know
58:07
me i don't know
58:07
me i don't know i've been very lucky to have a lot of
58:08
i've been very lucky to have a lot of
58:08
i've been very lucky to have a lot of toys to play with
58:10
toys to play with
58:10
toys to play with um the biggest and most impactful
58:13
um the biggest and most impactful
58:13
um the biggest and most impactful devices that i see
58:15
devices that i see
58:15
devices that i see are the ones that are readily accessible
58:17
are the ones that are readily accessible
58:17
are the ones that are readily accessible to the general masses
58:19
to the general masses
58:19
to the general masses i've been you know very lucky to have
58:21
i've been you know very lucky to have
58:21
i've been you know very lucky to have the opportunity to work at microsoft and
58:23
the opportunity to work at microsoft and
58:23
the opportunity to work at microsoft and have access to a hololens
58:25
have access to a hololens
58:25
have access to a hololens um i have a hololens gen one i don't
58:27
um i have a hololens gen one i don't
58:27
um i have a hololens gen one i don't have a gen two uh that i get to play
58:28
have a gen two uh that i get to play
58:28
have a gen two uh that i get to play with
58:29
with
58:29
with and the the while it's really cool
58:33
and the the while it's really cool
58:33
and the the while it's really cool you understand that you know the
58:34
you understand that you know the
58:34
you understand that you know the availability of a device like that
58:36
availability of a device like that
58:36
availability of a device like that at four to five thousand dollars it's
58:38
at four to five thousand dollars it's
58:38
at four to five thousand dollars it's it's you know it's an expensive device
58:40
it's you know it's an expensive device
58:40
it's you know it's an expensive device that's canadian i don't i don't know
58:41
that's canadian i don't i don't know
58:41
that's canadian i don't i don't know what the transportation would be into
58:43
what the transportation would be into
58:43
what the transportation would be into into us or other currencies the the big
58:46
into us or other currencies the the big
58:46
into us or other currencies the the big thing
58:46
thing
58:46
thing is the cardboard box and the
58:49
is the cardboard box and the
58:49
is the cardboard box and the availability
58:49
availability
58:50
availability that you know i've i've been to
58:52
that you know i've i've been to
58:52
that you know i've i've been to elementary schools
58:53
elementary schools
58:53
elementary schools here and have shared the cardboard box
58:55
here and have shared the cardboard box
58:56
here and have shared the cardboard box idea with minecraft
58:58
idea with minecraft
58:58
idea with minecraft and the kids all have phones like iphone
59:01
and the kids all have phones like iphone
59:01
and the kids all have phones like iphone 6
59:01
6
59:01
6 galaxy s4 you know phones that are four
59:04
galaxy s4 you know phones that are four
59:04
galaxy s4 you know phones that are four or five years old
59:05
or five years old
59:05
or five years old and having the ability to understand and
59:07
and having the ability to understand and
59:07
and having the ability to understand and utilize a device in another way that
59:09
utilize a device in another way that
59:09
utilize a device in another way that they don't
59:09
they don't
59:10
they don't know is available to them uh using you
59:13
know is available to them uh using you
59:13
know is available to them uh using you know
59:13
know
59:13
know you use the minecraft world technology
59:15
you use the minecraft world technology
59:15
you use the minecraft world technology and they're now building
59:16
and they're now building
59:16
and they're now building uh structures inside of their classroom
59:19
uh structures inside of their classroom
59:19
uh structures inside of their classroom and they make it functional
59:21
and they make it functional
59:21
and they make it functional and the teachers invite them to hey
59:22
and the teachers invite them to hey
59:22
and the teachers invite them to hey build out the study area using minecraft
59:25
build out the study area using minecraft
59:25
build out the study area using minecraft and so there's that type of interaction
59:27
and so there's that type of interaction
59:27
and so there's that type of interaction and you see their eyes light up and go
59:28
and you see their eyes light up and go
59:28
and you see their eyes light up and go wow that's amazing
59:31
wow that's amazing
59:31
wow that's amazing the likelihood of each student getting a
59:33
the likelihood of each student getting a
59:33
the likelihood of each student getting a homeless
59:34
homeless
59:34
homeless i would love to see that but you know
59:35
i would love to see that but you know
59:36
i would love to see that but you know it's very difficult the availability of
59:38
it's very difficult the availability of
59:38
it's very difficult the availability of the cardboard box
59:40
the cardboard box
59:40
the cardboard box makes it you know on 11 level playing
59:42
makes it you know on 11 level playing
59:42
makes it you know on 11 level playing field in terms of ideas and enablement
59:44
field in terms of ideas and enablement
59:44
field in terms of ideas and enablement with the hopes that you know down the
59:46
with the hopes that you know down the
59:46
with the hopes that you know down the road they reinvent what services
59:48
road they reinvent what services
59:48
road they reinvent what services microsoft need to
59:49
microsoft need to
59:49
microsoft need to need to create yeah
59:53
need to create yeah
59:53
need to create yeah yeah and i was thinking you mentioned
59:55
yeah and i was thinking you mentioned
59:55
yeah and i was thinking you mentioned minecraft
59:58
minecraft
59:58
minecraft so if people decide to
1:00:01
so if people decide to
1:00:01
so if people decide to take this on this sort of a project
1:00:03
take this on this sort of a project
1:00:03
take this on this sort of a project procedural generation of stuff in his
1:00:05
procedural generation of stuff in his
1:00:05
procedural generation of stuff in his space
1:00:06
space
1:00:06
space um would you what would you say is a
1:00:09
um would you what would you say is a
1:00:09
um would you what would you say is a good
1:00:09
good
1:00:10
good uh environment to start using ai
1:00:13
uh environment to start using ai
1:00:13
uh environment to start using ai in that case
1:00:16
i would actually before even dabbling in
1:00:18
i would actually before even dabbling in
1:00:18
i would actually before even dabbling in an environment i would actually
1:00:20
an environment i would actually
1:00:20
an environment i would actually fully map out what are you trying to
1:00:22
fully map out what are you trying to
1:00:22
fully map out what are you trying to address i think the
1:00:24
address i think the
1:00:24
address i think the the technology follows suit in regards
1:00:26
the technology follows suit in regards
1:00:26
the technology follows suit in regards to what
1:00:27
to what
1:00:27
to what devices you have at your disposal or
1:00:29
devices you have at your disposal or
1:00:29
devices you have at your disposal or what services you have at your disposal
1:00:31
what services you have at your disposal
1:00:31
what services you have at your disposal i think that's what you know obviously
1:00:33
i think that's what you know obviously
1:00:33
i think that's what you know obviously being comfortable with a specific
1:00:34
being comfortable with a specific
1:00:34
being comfortable with a specific platform or technology is also going to
1:00:36
platform or technology is also going to
1:00:36
platform or technology is also going to be very important in terms of your
1:00:37
be very important in terms of your
1:00:37
be very important in terms of your adoption
1:00:38
adoption
1:00:38
adoption but starting in terms of understanding
1:00:40
but starting in terms of understanding
1:00:40
but starting in terms of understanding what the problem exists first or the
1:00:41
what the problem exists first or the
1:00:41
what the problem exists first or the opportunity
1:00:42
opportunity
1:00:42
opportunity exists first in terms of what needs to
1:00:43
exists first in terms of what needs to
1:00:43
exists first in terms of what needs to be addressed and really detailing that
1:00:44
be addressed and really detailing that
1:00:44
be addressed and really detailing that out
1:00:45
out
1:00:45
out i can't stress that enough because i've
1:00:47
i can't stress that enough because i've
1:00:47
i can't stress that enough because i've seen it time and time again where
1:00:49
seen it time and time again where
1:00:49
seen it time and time again where people will lead with technology
1:00:50
people will lead with technology
1:00:50
people will lead with technology organizations will lead with technology
1:00:52
organizations will lead with technology
1:00:52
organizations will lead with technology and they'll go down the route and make
1:00:54
and they'll go down the route and make
1:00:54
and they'll go down the route and make this huge investment in terms of
1:00:55
this huge investment in terms of
1:00:55
this huge investment in terms of adoption of said technology
1:00:57
adoption of said technology
1:00:57
adoption of said technology not thinking about the limitations that
1:00:58
not thinking about the limitations that
1:00:58
not thinking about the limitations that may occur in terms of
1:01:00
may occur in terms of
1:01:00
may occur in terms of devices availability of services online
1:01:03
devices availability of services online
1:01:03
devices availability of services online right
1:01:03
right
1:01:03
right and then being stopped midway or having
1:01:06
and then being stopped midway or having
1:01:06
and then being stopped midway or having to
1:01:07
to
1:01:08
to change their you know requirements to a
1:01:10
change their you know requirements to a
1:01:10
change their you know requirements to a lesser requirement because they can't
1:01:12
lesser requirement because they can't
1:01:12
lesser requirement because they can't move forward unless they do so
1:01:13
move forward unless they do so
1:01:13
move forward unless they do so missing out on part of the opportunity
1:01:15
missing out on part of the opportunity
1:01:15
missing out on part of the opportunity right i think you know
1:01:16
right i think you know
1:01:16
right i think you know if you're talking about platform and
1:01:18
if you're talking about platform and
1:01:18
if you're talking about platform and enablement and language to code into
1:01:21
enablement and language to code into
1:01:21
enablement and language to code into you know a lot of the projects that you
1:01:22
you know a lot of the projects that you
1:01:22
you know a lot of the projects that you know we build we build in in such a way
1:01:25
know we build we build in in such a way
1:01:25
know we build we build in in such a way that they're
1:01:26
that they're
1:01:26
that they're malleable you can use them on any
1:01:28
malleable you can use them on any
1:01:28
malleable you can use them on any platform and that's why
1:01:29
platform and that's why
1:01:29
platform and that's why i love showcasing the drone solution
1:01:31
i love showcasing the drone solution
1:01:31
i love showcasing the drone solution because it's evolved into so many things
1:01:34
because it's evolved into so many things
1:01:34
because it's evolved into so many things and i'll be honest like the drone
1:01:35
and i'll be honest like the drone
1:01:36
and i'll be honest like the drone solution itself came from
1:01:37
solution itself came from
1:01:38
solution itself came from an accident in terms of myself
1:01:40
an accident in terms of myself
1:01:40
an accident in terms of myself connecting a raspberry pi
1:01:41
connecting a raspberry pi
1:01:41
connecting a raspberry pi to a mouse trap and understanding the
1:01:44
to a mouse trap and understanding the
1:01:44
to a mouse trap and understanding the the
1:01:45
the
1:01:45
the travel patterns of mice inside of
1:01:46
travel patterns of mice inside of
1:01:46
travel patterns of mice inside of warehouses and and that was from an iot
1:01:49
warehouses and and that was from an iot
1:01:49
warehouses and and that was from an iot experiment to see if we could even catch
1:01:51
experiment to see if we could even catch
1:01:51
experiment to see if we could even catch more mice doing that
1:01:52
more mice doing that
1:01:52
more mice doing that and the learning capability that came
1:01:54
and the learning capability that came
1:01:54
and the learning capability that came out of that in terms of machine learning
1:01:56
out of that in terms of machine learning
1:01:56
out of that in terms of machine learning it just blew my mind and then that
1:01:58
it just blew my mind and then that
1:01:58
it just blew my mind and then that opened it like that's the thing
1:01:59
opened it like that's the thing
1:01:59
opened it like that's the thing it's it's the ability to play the
1:02:01
it's it's the ability to play the
1:02:02
it's it's the ability to play the ability to understand the opportunity
1:02:03
ability to understand the opportunity
1:02:03
ability to understand the opportunity the ability to grow from that uh but
1:02:06
the ability to grow from that uh but
1:02:06
the ability to grow from that uh but doing so with what you have
1:02:07
doing so with what you have
1:02:07
doing so with what you have at your disposal not worrying about i
1:02:10
at your disposal not worrying about i
1:02:10
at your disposal not worrying about i have to go out and buy this five
1:02:11
have to go out and buy this five
1:02:11
have to go out and buy this five thousand dollar unit or else this
1:02:12
thousand dollar unit or else this
1:02:12
thousand dollar unit or else this solution is not gonna work
1:02:13
solution is not gonna work
1:02:13
solution is not gonna work never limit yourself by that right
1:02:16
never limit yourself by that right
1:02:16
never limit yourself by that right you're always able to
1:02:17
you're always able to
1:02:17
you're always able to move forward in terms of your idea as
1:02:19
move forward in terms of your idea as
1:02:20
move forward in terms of your idea as long as you understand
1:02:21
long as you understand
1:02:21
long as you understand what your idea is and what you have at
1:02:23
what your idea is and what you have at
1:02:23
what your idea is and what you have at your disposal and what you can create
1:02:25
your disposal and what you can create
1:02:25
your disposal and what you can create and you can still go out and ask and say
1:02:27
and you can still go out and ask and say
1:02:27
and you can still go out and ask and say you know are these services ever going
1:02:28
you know are these services ever going
1:02:28
you know are these services ever going to be available is this something
1:02:29
to be available is this something
1:02:29
to be available is this something you know the community like so important
1:02:32
you know the community like so important
1:02:32
you know the community like so important like you know
1:02:33
like you know
1:02:33
like you know this community here in terms of
1:02:34
this community here in terms of
1:02:34
this community here in terms of everybody that's participating online
1:02:36
everybody that's participating online
1:02:36
everybody that's participating online everybody has an idea everybody has a
1:02:38
everybody has an idea everybody has a
1:02:38
everybody has an idea everybody has a device or a service that they can bring
1:02:40
device or a service that they can bring
1:02:40
device or a service that they can bring into
1:02:40
into
1:02:40
into into the fold and share you know
1:02:42
into the fold and share you know
1:02:42
into the fold and share you know knowledge uh being made available to
1:02:44
knowledge uh being made available to
1:02:44
knowledge uh being made available to everybody to
1:02:45
everybody to
1:02:45
everybody to to move forward with huge opportunities
1:02:48
to move forward with huge opportunities
1:02:48
to move forward with huge opportunities in terms of growth in terms of
1:02:50
in terms of growth in terms of
1:02:50
in terms of growth in terms of creation in terms of you know addressing
1:02:51
creation in terms of you know addressing
1:02:51
creation in terms of you know addressing opportunities as well
1:02:55
yeah what about you fergus any
1:02:57
yeah what about you fergus any
1:02:57
yeah what about you fergus any recommendations for people
1:02:59
recommendations for people
1:03:00
recommendations for people how to approach uh such a complex
1:03:01
how to approach uh such a complex
1:03:02
how to approach uh such a complex project as
1:03:03
project as
1:03:03
project as as you worked on
1:03:06
um i would say you know to
1:03:09
um i would say you know to
1:03:10
um i would say you know to echo anthony you've got to look at the
1:03:13
echo anthony you've got to look at the
1:03:13
echo anthony you've got to look at the end goal and then sort of
1:03:15
end goal and then sort of
1:03:15
end goal and then sort of decide what the best technology and the
1:03:18
decide what the best technology and the
1:03:18
decide what the best technology and the best
1:03:18
best
1:03:18
best uh approach for you is and you know just
1:03:22
uh approach for you is and you know just
1:03:22
uh approach for you is and you know just keep everything as agile as possible
1:03:23
keep everything as agile as possible
1:03:23
keep everything as agile as possible that's the
1:03:24
that's the
1:03:24
that's the the best advice i i can give you is like
1:03:26
the best advice i i can give you is like
1:03:26
the best advice i i can give you is like make things in small
1:03:27
make things in small
1:03:27
make things in small iterations you don't have to start big
1:03:30
iterations you don't have to start big
1:03:30
iterations you don't have to start big you don't
1:03:30
you don't
1:03:30
you don't even if you have grand ambitions for a
1:03:33
even if you have grand ambitions for a
1:03:33
even if you have grand ambitions for a 3d
1:03:34
3d
1:03:34
3d uh visual space just download unity and
1:03:37
uh visual space just download unity and
1:03:37
uh visual space just download unity and put it
1:03:37
put it
1:03:37
put it put a cube in there and you know have
1:03:40
put a cube in there and you know have
1:03:40
put a cube in there and you know have have an explorer of your cube and then
1:03:41
have an explorer of your cube and then
1:03:41
have an explorer of your cube and then build up
1:03:42
build up
1:03:42
build up both your skills and your assets are
1:03:44
both your skills and your assets are
1:03:44
both your skills and your assets are sort of as you go and
1:03:46
sort of as you go and
1:03:46
sort of as you go and so don't be afraid to jump in and dive
1:03:48
so don't be afraid to jump in and dive
1:03:48
so don't be afraid to jump in and dive in in terms of
1:03:49
in in terms of
1:03:50
in in terms of um technology again agree that the the
1:03:53
um technology again agree that the the
1:03:53
um technology again agree that the the more and more accessible
1:03:54
more and more accessible
1:03:54
more and more accessible this becomes the easier it's going to be
1:03:56
this becomes the easier it's going to be
1:03:56
this becomes the easier it's going to be i mean if you went back to
1:03:58
i mean if you went back to
1:03:58
i mean if you went back to the 1980s and said you know every child
1:04:01
the 1980s and said you know every child
1:04:01
the 1980s and said you know every child is going to be required to have access
1:04:03
is going to be required to have access
1:04:03
is going to be required to have access to
1:04:03
to
1:04:03
to a computer to attend school one day
1:04:07
a computer to attend school one day
1:04:07
a computer to attend school one day um they would have thought you were mad
1:04:09
um they would have thought you were mad
1:04:09
um they would have thought you were mad because you know computers
1:04:11
because you know computers
1:04:11
because you know computers powerful computers with the size of
1:04:13
powerful computers with the size of
1:04:13
powerful computers with the size of rooms and people just didn't have
1:04:15
rooms and people just didn't have
1:04:15
rooms and people just didn't have uh an ability to afford that so as these
1:04:17
uh an ability to afford that so as these
1:04:18
uh an ability to afford that so as these things come down in price it's going to
1:04:19
things come down in price it's going to
1:04:19
things come down in price it's going to be so much more accessible
1:04:20
be so much more accessible
1:04:20
be so much more accessible and so much easier to get started with
1:04:22
and so much easier to get started with
1:04:22
and so much easier to get started with but crucially on the flip side is that
1:04:24
but crucially on the flip side is that
1:04:24
but crucially on the flip side is that we're going to need
1:04:25
we're going to need
1:04:25
we're going to need the skills and the the people to develop
1:04:28
the skills and the the people to develop
1:04:28
the skills and the the people to develop the new applications in the new
1:04:29
the new applications in the new
1:04:29
the new applications in the new environments
1:04:31
environments
1:04:31
environments as well so i'd say you know get learning
1:04:34
as well so i'd say you know get learning
1:04:34
as well so i'd say you know get learning get playing yeah and speaking of good
1:04:37
get playing yeah and speaking of good
1:04:37
get playing yeah and speaking of good resources
1:04:39
resources
1:04:39
resources to learn and play any recommendations
1:04:41
to learn and play any recommendations
1:04:41
to learn and play any recommendations for people
1:04:43
for people
1:04:43
for people maybe websites they can go to books they
1:04:46
maybe websites they can go to books they
1:04:46
maybe websites they can go to books they can read to to learn
1:04:48
can read to to learn
1:04:48
can read to to learn uh to either to work in a 3d space using
1:04:51
uh to either to work in a 3d space using
1:04:51
uh to either to work in a 3d space using virtual reality and ai
1:04:52
virtual reality and ai
1:04:52
virtual reality and ai and or flying a drone for that matter
1:04:55
and or flying a drone for that matter
1:04:55
and or flying a drone for that matter anthony any ideas
1:04:58
anthony any ideas
1:04:58
anthony any ideas actually i think he was going to say him
1:05:01
actually i think he was going to say him
1:05:01
actually i think he was going to say him first he's he's kind of he's got
1:05:02
first he's he's kind of he's got
1:05:02
first he's he's kind of he's got something to share i think
1:05:05
something to share i think
1:05:05
something to share i think i was just going to say like and there's
1:05:07
i was just going to say like and there's
1:05:07
i was just going to say like and there's tons of courses that you can pay for
1:05:08
tons of courses that you can pay for
1:05:08
tons of courses that you can pay for online
1:05:09
online
1:05:09
online which are pretty good but actually
1:05:11
which are pretty good but actually
1:05:11
which are pretty good but actually youtube is
1:05:12
youtube is
1:05:12
youtube is um a great starting place there's tons
1:05:14
um a great starting place there's tons
1:05:14
um a great starting place there's tons of like bite size
1:05:16
of like bite size
1:05:16
of like bite size um tutorials on youtube that i've used
1:05:19
um tutorials on youtube that i've used
1:05:19
um tutorials on youtube that i've used for both unity and blender
1:05:21
for both unity and blender
1:05:21
for both unity and blender these are free software these free
1:05:23
these are free software these free
1:05:23
these are free software these free pieces of software when you're using
1:05:24
pieces of software when you're using
1:05:24
pieces of software when you're using them for the first time
1:05:25
them for the first time
1:05:25
them for the first time or you're not using them in like a
1:05:27
or you're not using them in like a
1:05:27
or you're not using them in like a professional context so download the
1:05:29
professional context so download the
1:05:29
professional context so download the tools have a go look at some youtube
1:05:30
tools have a go look at some youtube
1:05:30
tools have a go look at some youtube videos and get started
1:05:32
videos and get started
1:05:32
videos and get started especially for blender and unity some of
1:05:34
especially for blender and unity some of
1:05:34
especially for blender and unity some of them are really nice bite-sized
1:05:36
them are really nice bite-sized
1:05:36
them are really nice bite-sized chunks which you can learn to do like
1:05:38
chunks which you can learn to do like
1:05:38
chunks which you can learn to do like one or two things step by step and then
1:05:40
one or two things step by step and then
1:05:40
one or two things step by step and then you start getting an idea and you can
1:05:42
you start getting an idea and you can
1:05:42
you start getting an idea and you can sort of go on and and then once you're
1:05:44
sort of go on and and then once you're
1:05:44
sort of go on and and then once you're ready i would say just challenge
1:05:46
ready i would say just challenge
1:05:46
ready i would say just challenge yourself come up with an idea of
1:05:47
yourself come up with an idea of
1:05:47
yourself come up with an idea of something you would like to 3d model or
1:05:48
something you would like to 3d model or
1:05:48
something you would like to 3d model or an environment you'd like to create
1:05:50
an environment you'd like to create
1:05:50
an environment you'd like to create and have a go start basic and have a go
1:05:52
and have a go start basic and have a go
1:05:52
and have a go start basic and have a go in terms of flying drones i have some
1:05:54
in terms of flying drones i have some
1:05:54
in terms of flying drones i have some limited experience with that but i would
1:05:56
limited experience with that but i would
1:05:56
limited experience with that but i would say
1:05:57
say
1:05:57
say invest in a drone that can fly itself if
1:05:58
invest in a drone that can fly itself if
1:05:58
invest in a drone that can fly itself if you are not as coordinated as i am
1:06:01
you are not as coordinated as i am
1:06:01
you are not as coordinated as i am or not no as uncoordinated as i am
1:06:04
or not no as uncoordinated as i am
1:06:04
or not no as uncoordinated as i am i've definitely crashed a couple of them
1:06:07
i've definitely crashed a couple of them
1:06:08
i've definitely crashed a couple of them cool actually all the ones behind me are
1:06:09
cool actually all the ones behind me are
1:06:09
cool actually all the ones behind me are broken that's why they're up there
1:06:11
broken that's why they're up there
1:06:12
broken that's why they're up there how many how many drones did you break
1:06:13
how many how many drones did you break
1:06:13
how many how many drones did you break anthony oh too many
1:06:15
anthony oh too many
1:06:15
anthony oh too many two my drones are now made of foam so
1:06:18
two my drones are now made of foam so
1:06:18
two my drones are now made of foam so so they don't break um i you know again
1:06:21
so they don't break um i you know again
1:06:21
so they don't break um i you know again i
1:06:22
i
1:06:22
i i echo uh in terms of the the youtube
1:06:25
i echo uh in terms of the the youtube
1:06:25
i echo uh in terms of the the youtube uh and the videos that are available
1:06:26
uh and the videos that are available
1:06:26
uh and the videos that are available online um i like to get my hands dirty i
1:06:29
online um i like to get my hands dirty i
1:06:29
online um i like to get my hands dirty i like to play
1:06:30
like to play
1:06:30
like to play uh directly in uh environments um
1:06:33
uh directly in uh environments um
1:06:33
uh directly in uh environments um there is you know sometimes the cost
1:06:35
there is you know sometimes the cost
1:06:35
there is you know sometimes the cost that you have to adhere to in regards to
1:06:37
that you have to adhere to in regards to
1:06:37
that you have to adhere to in regards to that
1:06:37
that
1:06:37
that um microsoft has a phenomenal solution
1:06:40
um microsoft has a phenomenal solution
1:06:40
um microsoft has a phenomenal solution called microsoft learn
1:06:41
called microsoft learn
1:06:41
called microsoft learn uh it's i believe it's microsoft.com
1:06:43
uh it's i believe it's microsoft.com
1:06:43
uh it's i believe it's microsoft.com forward slash learn
1:06:45
forward slash learn
1:06:45
forward slash learn they've invoked what we call sandbox
1:06:47
they've invoked what we call sandbox
1:06:47
they've invoked what we call sandbox technology which allows you
1:06:49
technology which allows you
1:06:49
technology which allows you at no cost to dabble in azure
1:06:51
at no cost to dabble in azure
1:06:51
at no cost to dabble in azure environments uh
1:06:52
environments uh
1:06:52
environments uh to build out solutions and test out
1:06:54
to build out solutions and test out
1:06:54
to build out solutions and test out theories uh and
1:06:55
theories uh and
1:06:55
theories uh and provide you step-by-step instruction to
1:06:57
provide you step-by-step instruction to
1:06:57
provide you step-by-step instruction to do so i do know that there
1:06:59
do so i do know that there
1:06:59
do so i do know that there are uh some augmented reality uh learn
1:07:02
are uh some augmented reality uh learn
1:07:02
are uh some augmented reality uh learn modules that are out there
1:07:03
modules that are out there
1:07:03
modules that are out there and if there's a specific module uh that
1:07:06
and if there's a specific module uh that
1:07:06
and if there's a specific module uh that is
1:07:06
is
1:07:06
is uh a requirement for yourself in terms
1:07:08
uh a requirement for yourself in terms
1:07:08
uh a requirement for yourself in terms of upskilling is not there
1:07:10
of upskilling is not there
1:07:10
of upskilling is not there let me know i'm directly tuned in with
1:07:12
let me know i'm directly tuned in with
1:07:12
let me know i'm directly tuned in with the ms learn team and
1:07:13
the ms learn team and
1:07:13
the ms learn team and we're actively building out new modules
1:07:15
we're actively building out new modules
1:07:15
we're actively building out new modules as we speak
1:07:17
as we speak
1:07:17
as we speak based on your requirements based on you
1:07:19
based on your requirements based on you
1:07:19
based on your requirements based on you know your audience's requirements
1:07:20
know your audience's requirements
1:07:20
know your audience's requirements in terms of what they would like to see
1:07:22
in terms of what they would like to see
1:07:22
in terms of what they would like to see further you know learnings on or
1:07:24
further you know learnings on or
1:07:24
further you know learnings on or interactions in
1:07:25
interactions in
1:07:25
interactions in and stuff like the drone solution that
1:07:27
and stuff like the drone solution that
1:07:28
and stuff like the drone solution that we'll be talking about later
1:07:29
we'll be talking about later
1:07:29
we'll be talking about later we've incorporated that into learning as
1:07:31
we've incorporated that into learning as
1:07:31
we've incorporated that into learning as a real world story to see how
1:07:33
a real world story to see how
1:07:33
a real world story to see how you know creativity comes into play to
1:07:35
you know creativity comes into play to
1:07:35
you know creativity comes into play to address opportunities with technology
1:07:37
address opportunities with technology
1:07:37
address opportunities with technology but not leave with technology to to
1:07:40
but not leave with technology to to
1:07:40
but not leave with technology to to create an opportunity
1:07:42
create an opportunity
1:07:42
create an opportunity uh so that one is an important one and
1:07:44
uh so that one is an important one and
1:07:44
uh so that one is an important one and then last but not least is the community
1:07:46
then last but not least is the community
1:07:46
then last but not least is the community i can't tell you how much i've learned
1:07:48
i can't tell you how much i've learned
1:07:48
i can't tell you how much i've learned from the community in terms of
1:07:50
from the community in terms of
1:07:50
from the community in terms of what you know what people bring to the
1:07:52
what you know what people bring to the
1:07:52
what you know what people bring to the table everybody has a different you know
1:07:53
table everybody has a different you know
1:07:53
table everybody has a different you know experience everybody has it you know
1:07:55
experience everybody has it you know
1:07:55
experience everybody has it you know comes from a different walk of life
1:07:57
comes from a different walk of life
1:07:57
comes from a different walk of life i you know started my career as a car
1:07:58
i you know started my career as a car
1:07:58
i you know started my career as a car mechanic i was you know
1:08:00
mechanic i was you know
1:08:00
mechanic i was you know my my first uh jump into technology was
1:08:03
my my first uh jump into technology was
1:08:03
my my first uh jump into technology was when they
1:08:04
when they
1:08:04
when they introduced the ecu uh into the vehicles
1:08:06
introduced the ecu uh into the vehicles
1:08:06
introduced the ecu uh into the vehicles and i was connecting the
1:08:07
and i was connecting the
1:08:07
and i was connecting the serial port on a 486 to an obd2
1:08:10
serial port on a 486 to an obd2
1:08:10
serial port on a 486 to an obd2 connector
1:08:11
connector
1:08:11
connector and extracting data from these vehicles
1:08:13
and extracting data from these vehicles
1:08:13
and extracting data from these vehicles in terms of what's the problem
1:08:14
in terms of what's the problem
1:08:14
in terms of what's the problem and how to fix them and i just this was
1:08:16
and how to fix them and i just this was
1:08:16
and how to fix them and i just this was amazing and it was that's
1:08:19
amazing and it was that's
1:08:19
amazing and it was that's not even ai not even machine learning at
1:08:21
not even ai not even machine learning at
1:08:21
not even ai not even machine learning at that time was just literally
1:08:22
that time was just literally
1:08:22
that time was just literally understanding here's the problem sheet
1:08:24
understanding here's the problem sheet
1:08:24
understanding here's the problem sheet and knowing by the error codes that was
1:08:25
and knowing by the error codes that was
1:08:25
and knowing by the error codes that was coming out from this
1:08:26
coming out from this
1:08:26
coming out from this from this car's computer what i needed
1:08:28
from this car's computer what i needed
1:08:28
from this car's computer what i needed to fix it when i needed to address and
1:08:30
to fix it when i needed to address and
1:08:30
to fix it when i needed to address and that's where i moved into into my career
1:08:31
that's where i moved into into my career
1:08:31
that's where i moved into into my career into technology
1:08:32
into technology
1:08:32
into technology um you know bringing that to the table
1:08:35
um you know bringing that to the table
1:08:35
um you know bringing that to the table and
1:08:36
and
1:08:36
and addressing a problem first with before
1:08:38
addressing a problem first with before
1:08:38
addressing a problem first with before even incorporating technologies what i
1:08:40
even incorporating technologies what i
1:08:40
even incorporating technologies what i brought with me from
1:08:41
brought with me from
1:08:41
brought with me from from my experience as a mechanic other
1:08:43
from my experience as a mechanic other
1:08:43
from my experience as a mechanic other people bring other experiences
1:08:45
people bring other experiences
1:08:45
people bring other experiences the one advantage i've had in terms of
1:08:47
the one advantage i've had in terms of
1:08:47
the one advantage i've had in terms of traveling around the world is listening
1:08:49
traveling around the world is listening
1:08:49
traveling around the world is listening to people
1:08:50
to people
1:08:50
to people in terms of how they address
1:08:51
in terms of how they address
1:08:51
in terms of how they address opportunities and experiences everywhere
1:08:53
opportunities and experiences everywhere
1:08:53
opportunities and experiences everywhere from europe to africa to australia
1:08:55
from europe to africa to australia
1:08:55
from europe to africa to australia you name it and everybody addresses you
1:08:58
you name it and everybody addresses you
1:08:58
you name it and everybody addresses you know problems and
1:08:59
know problems and
1:08:59
know problems and and opportunities differently community
1:09:02
and opportunities differently community
1:09:02
and opportunities differently community is so important to to have that
1:09:04
is so important to to have that
1:09:04
is so important to to have that capability to listen and it misses
1:09:06
capability to listen and it misses
1:09:06
capability to listen and it misses pandemic
1:09:07
pandemic
1:09:07
pandemic uh it's been even more beneficial in
1:09:09
uh it's been even more beneficial in
1:09:09
uh it's been even more beneficial in regards to the community coming together
1:09:11
regards to the community coming together
1:09:11
regards to the community coming together because
1:09:12
because
1:09:12
because with us all being online we're all
1:09:14
with us all being online we're all
1:09:14
with us all being online we're all connected
1:09:15
connected
1:09:15
connected and yes i miss the personal touch i miss
1:09:17
and yes i miss the personal touch i miss
1:09:17
and yes i miss the personal touch i miss the handshakes and
1:09:18
the handshakes and
1:09:18
the handshakes and and the high fives and talking to people
1:09:21
and the high fives and talking to people
1:09:21
and the high fives and talking to people face to face
1:09:22
face to face
1:09:22
face to face but i've sat in a whole bunch of user
1:09:24
but i've sat in a whole bunch of user
1:09:24
but i've sat in a whole bunch of user group meetings from around the world
1:09:26
group meetings from around the world
1:09:26
group meetings from around the world uh that i wouldn't have had access to
1:09:28
uh that i wouldn't have had access to
1:09:28
uh that i wouldn't have had access to prior to this all going on
1:09:30
prior to this all going on
1:09:30
prior to this all going on and you know you sit in a session in
1:09:32
and you know you sit in a session in
1:09:32
and you know you sit in a session in india and it's in the session in
1:09:33
india and it's in the session in
1:09:33
india and it's in the session in australia
1:09:34
australia
1:09:34
australia and you know even if they're not in
1:09:36
and you know even if they're not in
1:09:36
and you know even if they're not in english there's a technology that's
1:09:38
english there's a technology that's
1:09:38
english there's a technology that's available that'll translate it for you
1:09:39
available that'll translate it for you
1:09:40
available that'll translate it for you uh just go out there participate you
1:09:42
uh just go out there participate you
1:09:42
uh just go out there participate you know
1:09:43
know
1:09:43
know not you're not just you don't just speak
1:09:45
not you're not just you don't just speak
1:09:45
not you're not just you don't just speak like i try to practice
1:09:46
like i try to practice
1:09:46
like i try to practice you know 20 speaking 80 percent
1:09:48
you know 20 speaking 80 percent
1:09:48
you know 20 speaking 80 percent listening
1:09:49
listening
1:09:49
listening you know take in what you're learning
1:09:51
you know take in what you're learning
1:09:51
you know take in what you're learning take in what you're you're seeing
1:09:52
take in what you're you're seeing
1:09:52
take in what you're you're seeing take notes take you know also share but
1:09:55
take notes take you know also share but
1:09:55
take notes take you know also share but do a lot of listening do a lot of
1:09:57
do a lot of listening do a lot of
1:09:57
do a lot of listening do a lot of understanding see what's going on around
1:09:58
understanding see what's going on around
1:09:58
understanding see what's going on around the world
1:09:59
the world
1:09:59
the world and you'll be amazed how much you'll
1:10:00
and you'll be amazed how much you'll
1:10:00
and you'll be amazed how much you'll learn you know from other people
1:10:02
learn you know from other people
1:10:02
learn you know from other people yeah yeah that's the reason the exact
1:10:05
yeah yeah that's the reason the exact
1:10:05
yeah yeah that's the reason the exact reason why we run the global ai october
1:10:07
reason why we run the global ai october
1:10:08
reason why we run the global ai october sessions we want to connect people
1:10:09
sessions we want to connect people
1:10:09
sessions we want to connect people across the globe because we feel that
1:10:12
across the globe because we feel that
1:10:12
across the globe because we feel that this is one of the ways that we can help
1:10:14
this is one of the ways that we can help
1:10:14
this is one of the ways that we can help people
1:10:15
people
1:10:16
people get new ideas actually for new projects
1:10:18
get new ideas actually for new projects
1:10:18
get new ideas actually for new projects so it's it's really cool to hear about
1:10:20
so it's it's really cool to hear about
1:10:20
so it's it's really cool to hear about the ideas that you have fergus
1:10:22
the ideas that you have fergus
1:10:22
the ideas that you have fergus and i want to thank you for your time
1:10:24
and i want to thank you for your time
1:10:24
and i want to thank you for your time today to show us actually what you built
1:10:27
today to show us actually what you built
1:10:27
today to show us actually what you built and if anybody has any more questions
1:10:29
and if anybody has any more questions
1:10:29
and if anybody has any more questions for fergus
1:10:30
for fergus
1:10:30
for fergus please feel free to post them in the
1:10:32
please feel free to post them in the
1:10:32
please feel free to post them in the live chat i guess fergus can stick
1:10:34
live chat i guess fergus can stick
1:10:34
live chat i guess fergus can stick around in the live chat and have a look
1:10:36
around in the live chat and have a look
1:10:36
around in the live chat and have a look at that
1:10:36
at that
1:10:36
at that um yeah and alicia i guess that we
1:10:39
um yeah and alicia i guess that we
1:10:39
um yeah and alicia i guess that we should definitely have a talk after the
1:10:41
should definitely have a talk after the
1:10:41
should definitely have a talk after the october sessions
1:10:42
october sessions
1:10:42
october sessions and and i have a bit of rest um
1:10:45
and and i have a bit of rest um
1:10:46
and and i have a bit of rest um getting back from this and we should
1:10:47
getting back from this and we should
1:10:47
getting back from this and we should definitely
1:10:49
definitely
1:10:49
definitely give this a shot and see what we can
1:10:50
give this a shot and see what we can
1:10:50
give this a shot and see what we can make i'm excited at least
1:10:56
yes i we have um our
1:11:00
yes i we have um our
1:11:00
yes i we have um our boot counts coming up in february and um
1:11:03
boot counts coming up in february and um
1:11:03
boot counts coming up in february and um we'd love to have both of you
1:11:04
we'd love to have both of you
1:11:04
we'd love to have both of you involved of course and uh
1:11:07
involved of course and uh
1:11:08
involved of course and uh definitely it's it's kind of heartening
1:11:10
definitely it's it's kind of heartening
1:11:10
definitely it's it's kind of heartening to hear that you know the genesis for
1:11:13
to hear that you know the genesis for
1:11:13
to hear that you know the genesis for for some of these projects are building
1:11:15
for some of these projects are building
1:11:15
for some of these projects are building a better mousetrap
1:11:16
a better mousetrap
1:11:16
a better mousetrap like i i think you literally said that
1:11:18
like i i think you literally said that
1:11:18
like i i think you literally said that and
1:11:19
and
1:11:19
and i think we all feel that you know
1:11:23
i think we all feel that you know
1:11:23
i think we all feel that you know the mousetrap has already been invented
1:11:26
the mousetrap has already been invented
1:11:26
the mousetrap has already been invented and
1:11:26
and
1:11:26
and there's nothing there's not much we can
1:11:28
there's nothing there's not much we can
1:11:28
there's nothing there's not much we can do to make things better
1:11:31
do to make things better
1:11:31
do to make things better but um you know there's always an
1:11:33
but um you know there's always an
1:11:33
but um you know there's always an opportunity
1:11:34
opportunity
1:11:34
opportunity to to build something new so thank you
1:11:36
to to build something new so thank you
1:11:36
to to build something new so thank you anthony for that and
1:11:37
anthony for that and
1:11:37
anthony for that and fergus it was great having wine yeah
1:11:40
fergus it was great having wine yeah
1:11:40
fergus it was great having wine yeah thanks so much for inviting the
1:11:42
thanks so much for inviting the
1:11:42
thanks so much for inviting the opportunity to speak to you all hope you
1:11:43
opportunity to speak to you all hope you
1:11:43
opportunity to speak to you all hope you found it uh interesting
1:11:45
found it uh interesting
1:11:45
found it uh interesting yeah thank you absolutely hope to see
1:11:47
yeah thank you absolutely hope to see
1:11:47
yeah thank you absolutely hope to see you soon
1:11:49
you soon
1:11:49
you soon yeah definitely thanks very much so
1:11:52
yeah definitely thanks very much so
1:11:52
yeah definitely thanks very much so um moving on to anthony um
1:11:56
um moving on to anthony um
1:11:56
um moving on to anthony um uh you've been on our show before i
1:11:58
uh you've been on our show before i
1:11:58
uh you've been on our show before i remember
1:11:59
remember
1:11:59
remember in april actually i don't know if i was
1:12:02
in april actually i don't know if i was
1:12:02
in april actually i don't know if i was asleep during your session
1:12:04
asleep during your session
1:12:04
asleep during your session or maybe i forgot because it was a
1:12:06
or maybe i forgot because it was a
1:12:06
or maybe i forgot because it was a 24-hour thing uh i don't know about you
1:12:08
24-hour thing uh i don't know about you
1:12:08
24-hour thing uh i don't know about you alicia
1:12:09
alicia
1:12:09
alicia do you remember this i remember anthony
1:12:12
do you remember this i remember anthony
1:12:12
do you remember this i remember anthony from our
1:12:13
from our
1:12:13
from our our 24-hour run uh 36-hour run
1:12:17
our 24-hour run uh 36-hour run
1:12:17
our 24-hour run uh 36-hour run and um i i really enjoyed
1:12:20
and um i i really enjoyed
1:12:20
and um i i really enjoyed your blending on the technologies and
1:12:23
your blending on the technologies and
1:12:23
your blending on the technologies and i i love your spiritual community so
1:12:26
i i love your spiritual community so
1:12:26
i i love your spiritual community so thank you for visiting us again
1:12:28
thank you for visiting us again
1:12:28
thank you for visiting us again oh thank you for having me i i'm really
1:12:31
oh thank you for having me i i'm really
1:12:31
oh thank you for having me i i'm really curious about
1:12:32
curious about
1:12:32
curious about about the project that you're going to
1:12:33
about the project that you're going to
1:12:33
about the project that you're going to talk to us
1:12:35
talk to us
1:12:35
talk to us about um i know you've you've just shown
1:12:38
about um i know you've you've just shown
1:12:38
about um i know you've you've just shown that the mousetrap thing uh
1:12:40
that the mousetrap thing uh
1:12:40
that the mousetrap thing uh uh behind you um and you've done a lot
1:12:42
uh behind you um and you've done a lot
1:12:42
uh behind you um and you've done a lot of stuff about
1:12:43
of stuff about
1:12:43
of stuff about with drones can you explain to us a
1:12:45
with drones can you explain to us a
1:12:45
with drones can you explain to us a little bit what you've done with drones
1:12:47
little bit what you've done with drones
1:12:47
little bit what you've done with drones and
1:12:48
and
1:12:48
and and how that works actually yeah um
1:12:50
and how that works actually yeah um
1:12:50
and how that works actually yeah um actually
1:12:51
actually
1:12:51
actually what i can do is i'm going to quickly
1:12:54
what i can do is i'm going to quickly
1:12:54
what i can do is i'm going to quickly share
1:12:55
share
1:12:55
share my screen i don't i don't want to go
1:12:56
my screen i don't i don't want to go
1:12:56
my screen i don't i don't want to go through a full-on powerpoint
1:12:58
through a full-on powerpoint
1:12:58
through a full-on powerpoint presentation but i did want to show
1:13:00
presentation but i did want to show
1:13:00
presentation but i did want to show a little bit around the technology what
1:13:02
a little bit around the technology what
1:13:02
a little bit around the technology what incorporated uh
1:13:04
incorporated uh
1:13:04
incorporated uh what we built out for the solution as
1:13:06
what we built out for the solution as
1:13:06
what we built out for the solution as mentioned this was done in partnership
1:13:07
mentioned this was done in partnership
1:13:08
mentioned this was done in partnership with intro robotics
1:13:09
with intro robotics
1:13:09
with intro robotics i'll just share this really quickly
1:13:11
i'll just share this really quickly
1:13:11
i'll just share this really quickly right here so this was done in
1:13:12
right here so this was done in
1:13:12
right here so this was done in partnership with intro robotics
1:13:14
partnership with intro robotics
1:13:14
partnership with intro robotics and like i mentioned before the
1:13:15
and like i mentioned before the
1:13:16
and like i mentioned before the opportunity that existed was
1:13:18
opportunity that existed was
1:13:18
opportunity that existed was indra had come to us and said hey we're
1:13:20
indra had come to us and said hey we're
1:13:20
indra had come to us and said hey we're doing this on behalf of the canadian
1:13:21
doing this on behalf of the canadian
1:13:21
doing this on behalf of the canadian coast guard they operate drones to
1:13:23
coast guard they operate drones to
1:13:23
coast guard they operate drones to you know fly out to scenarios or
1:13:25
you know fly out to scenarios or
1:13:25
you know fly out to scenarios or situations where a ship is in distress
1:13:27
situations where a ship is in distress
1:13:27
situations where a ship is in distress and it takes film of the scenario and
1:13:30
and it takes film of the scenario and
1:13:30
and it takes film of the scenario and then phil and then flies back
1:13:32
then phil and then flies back
1:13:32
then phil and then flies back these drones are gas-powered drones that
1:13:33
these drones are gas-powered drones that
1:13:33
these drones are gas-powered drones that can you know fly out to three hours out
1:13:35
can you know fly out to three hours out
1:13:36
can you know fly out to three hours out of the sea
1:13:37
of the sea
1:13:37
of the sea survey the area uh with with aerial
1:13:39
survey the area uh with with aerial
1:13:40
survey the area uh with with aerial footage and then fly back
1:13:41
footage and then fly back
1:13:42
footage and then fly back and in flying back it then comes back to
1:13:44
and in flying back it then comes back to
1:13:44
and in flying back it then comes back to the central office the central office
1:13:45
the central office the central office
1:13:45
the central office the central office downloads the video from
1:13:47
downloads the video from
1:13:47
downloads the video from from the tape and then does the analysis
1:13:49
from the tape and then does the analysis
1:13:50
from the tape and then does the analysis of the you know the situation that's at
1:13:51
of the you know the situation that's at
1:13:51
of the you know the situation that's at hand
1:13:52
hand
1:13:52
hand and that could mean that it's going
1:13:53
and that could mean that it's going
1:13:53
and that could mean that it's going through numerous hours
1:13:55
through numerous hours
1:13:55
through numerous hours uh worth of video to do so and the
1:13:58
uh worth of video to do so and the
1:13:58
uh worth of video to do so and the challenge with that is you know
1:14:00
challenge with that is you know
1:14:00
challenge with that is you know in terms of a life and it's in distress
1:14:02
in terms of a life and it's in distress
1:14:02
in terms of a life and it's in distress every second counts
1:14:03
every second counts
1:14:03
every second counts and you know there's also you know the
1:14:06
and you know there's also you know the
1:14:06
and you know there's also you know the the um preparedness of those services
1:14:10
the um preparedness of those services
1:14:10
the um preparedness of those services that go out to those scenarios
1:14:12
that go out to those scenarios
1:14:12
that go out to those scenarios to address you know the problem that's
1:14:14
to address you know the problem that's
1:14:14
to address you know the problem that's at hand is if you go out with a ship
1:14:15
at hand is if you go out with a ship
1:14:15
at hand is if you go out with a ship that doesn't have enough medical
1:14:16
that doesn't have enough medical
1:14:16
that doesn't have enough medical equipment to address the amount of
1:14:18
equipment to address the amount of
1:14:18
equipment to address the amount of people that are in the water that's also
1:14:19
people that are in the water that's also
1:14:19
people that are in the water that's also another issue
1:14:20
another issue
1:14:20
another issue and you know this piece here as it was
1:14:24
and you know this piece here as it was
1:14:24
and you know this piece here as it was initially brought to us
1:14:25
initially brought to us
1:14:25
initially brought to us was based on a technology ask
1:14:28
was based on a technology ask
1:14:28
was based on a technology ask can you help us provide a compression
1:14:31
can you help us provide a compression
1:14:31
can you help us provide a compression agent so that we can stream
1:14:33
agent so that we can stream
1:14:33
agent so that we can stream the video over for a 14.4 kilobits
1:14:37
the video over for a 14.4 kilobits
1:14:37
the video over for a 14.4 kilobits connection back to our central office so
1:14:39
connection back to our central office so
1:14:39
connection back to our central office so we can be
1:14:40
we can be
1:14:40
we can be immediate in our response of a scenario
1:14:43
immediate in our response of a scenario
1:14:43
immediate in our response of a scenario the biggest challenge with that is that
1:14:45
the biggest challenge with that is that
1:14:45
the biggest challenge with that is that you know in terms of regulations for
1:14:46
you know in terms of regulations for
1:14:46
you know in terms of regulations for search and rescue as you probably
1:14:48
search and rescue as you probably
1:14:48
search and rescue as you probably guessed there's a a plethora of rules
1:14:51
guessed there's a a plethora of rules
1:14:51
guessed there's a a plethora of rules behind the rescue of people
1:14:53
behind the rescue of people
1:14:53
behind the rescue of people uh from the joan perspective and in
1:14:56
uh from the joan perspective and in
1:14:56
uh from the joan perspective and in regards to this you know the video
1:14:57
regards to this you know the video
1:14:57
regards to this you know the video couldn't have
1:14:58
couldn't have
1:14:58
couldn't have could could not be lower than uh 720p in
1:15:00
could could not be lower than uh 720p in
1:15:00
could could not be lower than uh 720p in its quality
1:15:01
its quality
1:15:01
its quality uh the video itself and the drone the
1:15:03
uh the video itself and the drone the
1:15:04
uh the video itself and the drone the drone had to fly within a specific
1:15:05
drone had to fly within a specific
1:15:05
drone had to fly within a specific airspace you can't just have it fly
1:15:07
airspace you can't just have it fly
1:15:07
airspace you can't just have it fly um you know anywhere it has to fly
1:15:09
um you know anywhere it has to fly
1:15:09
um you know anywhere it has to fly within a specific band of airspace
1:15:11
within a specific band of airspace
1:15:11
within a specific band of airspace and because of commercial airlines and
1:15:13
and because of commercial airlines and
1:15:13
and because of commercial airlines and what have you uh
1:15:14
what have you uh
1:15:14
what have you uh and so you know with all these factors
1:15:16
and so you know with all these factors
1:15:16
and so you know with all these factors that are put into play
1:15:17
that are put into play
1:15:17
that are put into play uh there was you know up to 5 000 hours
1:15:21
uh there was you know up to 5 000 hours
1:15:21
uh there was you know up to 5 000 hours put forth in terms of adhering to not
1:15:23
put forth in terms of adhering to not
1:15:24
put forth in terms of adhering to not only the regulations that are required
1:15:25
only the regulations that are required
1:15:25
only the regulations that are required but also the specifications of the drone
1:15:27
but also the specifications of the drone
1:15:27
but also the specifications of the drone that we were given to do this
1:15:29
that we were given to do this
1:15:29
that we were given to do this uh proof of concept to do the
1:15:31
uh proof of concept to do the
1:15:31
uh proof of concept to do the identification of the individual that's
1:15:33
identification of the individual that's
1:15:33
identification of the individual that's in the water
1:15:33
in the water
1:15:34
in the water uh and that's and that's shown by the
1:15:35
uh and that's and that's shown by the
1:15:35
uh and that's and that's shown by the examples that you see here so this is a
1:15:37
examples that you see here so this is a
1:15:37
examples that you see here so this is a 720p
1:15:38
720p
1:15:38
720p camera at 5000 feet we had to test it at
1:15:41
camera at 5000 feet we had to test it at
1:15:41
camera at 5000 feet we had to test it at different
1:15:41
different
1:15:41
different type of weather patterns sunny day
1:15:43
type of weather patterns sunny day
1:15:43
type of weather patterns sunny day cloudy day rainy day snowy day
1:15:45
cloudy day rainy day snowy day
1:15:45
cloudy day rainy day snowy day you name it uh and to to do that
1:15:47
you name it uh and to to do that
1:15:48
you name it uh and to to do that understanding of what it was looking at
1:15:49
understanding of what it was looking at
1:15:49
understanding of what it was looking at at the water
1:15:50
at the water
1:15:50
at the water even the the uh life jacket itself is it
1:15:53
even the the uh life jacket itself is it
1:15:53
even the the uh life jacket itself is it red is it orange what are we you know
1:15:55
red is it orange what are we you know
1:15:55
red is it orange what are we you know the whole aspect of even how the hue of
1:15:57
the whole aspect of even how the hue of
1:15:57
the whole aspect of even how the hue of color
1:15:58
color
1:15:58
color looks in different aspects of light or
1:15:59
looks in different aspects of light or
1:15:59
looks in different aspects of light or even in the dark is how is the drone
1:16:01
even in the dark is how is the drone
1:16:01
even in the dark is how is the drone gonna identify the life jacket when it
1:16:03
gonna identify the life jacket when it
1:16:03
gonna identify the life jacket when it start
1:16:04
start
1:16:04
start uh you know it there was numerous tests
1:16:06
uh you know it there was numerous tests
1:16:06
uh you know it there was numerous tests that had to occur
1:16:07
that had to occur
1:16:07
that had to occur in regards to this the one thing i love
1:16:10
in regards to this the one thing i love
1:16:10
in regards to this the one thing i love telling people too is that i'm not a
1:16:11
telling people too is that i'm not a
1:16:11
telling people too is that i'm not a developer
1:16:12
developer
1:16:12
developer uh i i did not sit down and write a line
1:16:15
uh i i did not sit down and write a line
1:16:15
uh i i did not sit down and write a line of code and i
1:16:15
of code and i
1:16:16
of code and i i was uh um you know
1:16:19
i was uh um you know
1:16:19
i was uh um you know grateful that i had this awesome team uh
1:16:21
grateful that i had this awesome team uh
1:16:21
grateful that i had this awesome team uh to work with
1:16:22
to work with
1:16:22
to work with and collaborate on in regards to this
1:16:23
and collaborate on in regards to this
1:16:24
and collaborate on in regards to this i'm an infrastructure
1:16:25
i'm an infrastructure
1:16:25
i'm an infrastructure professional i'm you know as what we
1:16:27
professional i'm you know as what we
1:16:28
professional i'm you know as what we call an it pro
1:16:29
call an it pro
1:16:29
call an it pro that architects out the solution in
1:16:31
that architects out the solution in
1:16:31
that architects out the solution in terms of the uh where the data flows who
1:16:33
terms of the uh where the data flows who
1:16:33
terms of the uh where the data flows who gains access to the data what security
1:16:35
gains access to the data what security
1:16:35
gains access to the data what security aspects are put into play
1:16:37
aspects are put into play
1:16:37
aspects are put into play but i also you know dab a little bit in
1:16:39
but i also you know dab a little bit in
1:16:39
but i also you know dab a little bit in terms of architecture of these type of
1:16:40
terms of architecture of these type of
1:16:40
terms of architecture of these type of solutions in terms of
1:16:42
solutions in terms of
1:16:42
solutions in terms of can we actually make this happen and
1:16:43
can we actually make this happen and
1:16:43
can we actually make this happen and what i meant you know i
1:16:45
what i meant you know i
1:16:45
what i meant you know i mentioned earlier on the technology that
1:16:47
mentioned earlier on the technology that
1:16:47
mentioned earlier on the technology that we showcased and we highlighted
1:16:49
we showcased and we highlighted
1:16:49
we showcased and we highlighted from this solution didn't even exist
1:16:52
from this solution didn't even exist
1:16:52
from this solution didn't even exist customvision.ai as a service
1:16:54
customvision.ai as a service
1:16:54
customvision.ai as a service wasn't even commercially available when
1:16:56
wasn't even commercially available when
1:16:56
wasn't even commercially available when we built this out and a lot of people
1:16:58
we built this out and a lot of people
1:16:58
we built this out and a lot of people looked at music
1:16:58
looked at music
1:16:58
looked at music you can't do that there's no way the
1:17:00
you can't do that there's no way the
1:17:00
you can't do that there's no way the technology you know is not
1:17:01
technology you know is not
1:17:02
technology you know is not available for you to do that we're still
1:17:03
available for you to do that we're still
1:17:03
available for you to do that we're still in testing and you know the engineering
1:17:06
in testing and you know the engineering
1:17:06
in testing and you know the engineering team said you know
1:17:06
team said you know
1:17:06
team said you know okay we'll we'll grant you access to the
1:17:08
okay we'll we'll grant you access to the
1:17:08
okay we'll we'll grant you access to the service we'll you know
1:17:10
service we'll you know
1:17:10
service we'll you know but you have to keep us unlocked you
1:17:11
but you have to keep us unlocked you
1:17:11
but you have to keep us unlocked you have to you know go through the whole
1:17:13
have to you know go through the whole
1:17:13
have to you know go through the whole scenario with us in regards to
1:17:14
scenario with us in regards to
1:17:14
scenario with us in regards to what you're trying to accomplish uh and
1:17:16
what you're trying to accomplish uh and
1:17:16
what you're trying to accomplish uh and they were blown away because
1:17:18
they were blown away because
1:17:18
they were blown away because you know it was five thousand hours
1:17:19
you know it was five thousand hours
1:17:19
you know it was five thousand hours worth of training but the
1:17:22
worth of training but the
1:17:22
worth of training but the building of the infrastructure building
1:17:23
building of the infrastructure building
1:17:23
building of the infrastructure building of the technology you know it was done
1:17:25
of the technology you know it was done
1:17:25
of the technology you know it was done in the span of five days
1:17:26
in the span of five days
1:17:26
in the span of five days and you know the proof of concept has
1:17:28
and you know the proof of concept has
1:17:28
and you know the proof of concept has evolved into so
1:17:30
evolved into so
1:17:30
evolved into so many different types of solutions now
1:17:31
many different types of solutions now
1:17:31
many different types of solutions now you know i've seen um this scenario in
1:17:33
you know i've seen um this scenario in
1:17:33
you know i've seen um this scenario in terms of
1:17:34
terms of
1:17:34
terms of having the drone understand you know
1:17:36
having the drone understand you know
1:17:36
having the drone understand you know structure
1:17:37
structure
1:17:37
structure structural integrity of buildings and
1:17:39
structural integrity of buildings and
1:17:39
structural integrity of buildings and understanding if welds
1:17:40
understanding if welds
1:17:40
understanding if welds uh were actually completed appropriately
1:17:43
uh were actually completed appropriately
1:17:43
uh were actually completed appropriately i've seen the solution
1:17:43
i've seen the solution
1:17:44
i've seen the solution of monitoring um the the um
1:17:47
of monitoring um the the um
1:17:47
of monitoring um the the um fields of crops to see if there's enough
1:17:49
fields of crops to see if there's enough
1:17:49
fields of crops to see if there's enough moisture available
1:17:50
moisture available
1:17:50
moisture available uh to specifically you know let those
1:17:52
uh to specifically you know let those
1:17:52
uh to specifically you know let those crops grow
1:17:53
crops grow
1:17:53
crops grow optimally you know and that's just from
1:17:55
optimally you know and that's just from
1:17:56
optimally you know and that's just from a drone perspective
1:17:57
a drone perspective
1:17:57
a drone perspective you know amidst the other toys that i
1:17:58
you know amidst the other toys that i
1:17:58
you know amidst the other toys that i have back here i have a nvidia jensen
1:18:01
have back here i have a nvidia jensen
1:18:01
have back here i have a nvidia jensen uh which is a you know portable gpu it's
1:18:03
uh which is a you know portable gpu it's
1:18:03
uh which is a you know portable gpu it's like a raspberry pi on steroids
1:18:05
like a raspberry pi on steroids
1:18:05
like a raspberry pi on steroids uh with a built-in gpu capability uh at
1:18:08
uh with a built-in gpu capability uh at
1:18:08
uh with a built-in gpu capability uh at a cost effective
1:18:09
a cost effective
1:18:09
a cost effective uh piece that i can add a 720p
1:18:12
uh piece that i can add a 720p
1:18:12
uh piece that i can add a 720p camera to and bolt it onto anything a
1:18:15
camera to and bolt it onto anything a
1:18:15
camera to and bolt it onto anything a vehicle you know a a light post
1:18:17
vehicle you know a a light post
1:18:18
vehicle you know a a light post whatever that may be for whatever
1:18:19
whatever that may be for whatever
1:18:19
whatever that may be for whatever operation i want to accomplish
1:18:21
operation i want to accomplish
1:18:21
operation i want to accomplish again using the technology and the
1:18:23
again using the technology and the
1:18:23
again using the technology and the architecture that we built out on the
1:18:25
architecture that we built out on the
1:18:25
architecture that we built out on the drone
1:18:26
drone
1:18:26
drone i include uh my urls with the solutions
1:18:29
i include uh my urls with the solutions
1:18:29
i include uh my urls with the solutions that
1:18:29
that
1:18:30
that that i was able to collaborate on with
1:18:32
that i was able to collaborate on with
1:18:32
that i was able to collaborate on with others
1:18:33
others
1:18:33
others to either share the story or to actually
1:18:35
to either share the story or to actually
1:18:36
to either share the story or to actually share the code and provide access to my
1:18:37
share the code and provide access to my
1:18:37
share the code and provide access to my github repo for the code that
1:18:39
github repo for the code that
1:18:39
github repo for the code that what we've accomplished um and and i
1:18:42
what we've accomplished um and and i
1:18:42
what we've accomplished um and and i love the fact that i'm able to
1:18:43
love the fact that i'm able to
1:18:44
love the fact that i'm able to work with the community and interact
1:18:45
work with the community and interact
1:18:45
work with the community and interact with the community on these ideas and
1:18:46
with the community on these ideas and
1:18:46
with the community on these ideas and these solutions and share
1:18:48
these solutions and share
1:18:48
these solutions and share what i've built in the hopes that other
1:18:50
what i've built in the hopes that other
1:18:50
what i've built in the hopes that other people share them with me
1:18:51
people share them with me
1:18:51
people share them with me what they've accomplished and you know
1:18:53
what they've accomplished and you know
1:18:53
what they've accomplished and you know we combine efforts and we combine
1:18:55
we combine efforts and we combine
1:18:55
we combine efforts and we combine learnings
1:18:55
learnings
1:18:55
learnings to build up something really cool and
1:18:57
to build up something really cool and
1:18:57
to build up something really cool and and great um
1:18:59
and great um
1:18:59
and great um so as an example you know we use this as
1:19:02
so as an example you know we use this as
1:19:02
so as an example you know we use this as a
1:19:02
a
1:19:02
a as a fictitious uh scenario but it's
1:19:05
as a fictitious uh scenario but it's
1:19:05
as a fictitious uh scenario but it's something that's real world and it's
1:19:06
something that's real world and it's
1:19:06
something that's real world and it's something where a lot of organizations
1:19:08
something where a lot of organizations
1:19:08
something where a lot of organizations are facing today and it's this pandemic
1:19:11
are facing today and it's this pandemic
1:19:11
are facing today and it's this pandemic uh
1:19:11
uh
1:19:11
uh there is you know limited number amount
1:19:13
there is you know limited number amount
1:19:13
there is you know limited number amount of people that are allowed in spaces
1:19:15
of people that are allowed in spaces
1:19:15
of people that are allowed in spaces because there's
1:19:15
because there's
1:19:15
because there's specific distancing that is required uh
1:19:18
specific distancing that is required uh
1:19:18
specific distancing that is required uh in regards to you know having even
1:19:19
in regards to you know having even
1:19:20
in regards to you know having even people staff
1:19:21
people staff
1:19:21
people staff retail locations or stores not only
1:19:23
retail locations or stores not only
1:19:23
retail locations or stores not only those that are the customers that are
1:19:25
those that are the customers that are
1:19:25
those that are the customers that are coming in to participate
1:19:26
coming in to participate
1:19:26
coming in to participate uh so this company which is tail and
1:19:27
uh so this company which is tail and
1:19:28
uh so this company which is tail and traders has this exact problem
1:19:30
traders has this exact problem
1:19:30
traders has this exact problem every six months they have to do a full
1:19:32
every six months they have to do a full
1:19:32
every six months they have to do a full count of inventory
1:19:34
count of inventory
1:19:34
count of inventory of all the products that they sell in a
1:19:36
of all the products that they sell in a
1:19:36
of all the products that they sell in a physical inventory count
1:19:38
physical inventory count
1:19:38
physical inventory count to have an understanding to see that the
1:19:39
to have an understanding to see that the
1:19:39
to have an understanding to see that the correlation between what's listed in the
1:19:41
correlation between what's listed in the
1:19:41
correlation between what's listed in the computer it's available for sale
1:19:42
computer it's available for sale
1:19:42
computer it's available for sale and what's actually on hand now
1:19:45
and what's actually on hand now
1:19:45
and what's actually on hand now this usually happens after hours it has
1:19:47
this usually happens after hours it has
1:19:47
this usually happens after hours it has to happen in a time between
1:19:50
to happen in a time between
1:19:50
to happen in a time between you know when the close the store closes
1:19:52
you know when the close the store closes
1:19:52
you know when the close the store closes till the next morning
1:19:53
till the next morning
1:19:53
till the next morning but in a situation like what we're in
1:19:55
but in a situation like what we're in
1:19:55
but in a situation like what we're in right now with the pandemic
1:19:57
right now with the pandemic
1:19:57
right now with the pandemic you can't have as many staff as you you
1:19:59
you can't have as many staff as you you
1:19:59
you can't have as many staff as you you would usually have to do this count so
1:20:00
would usually have to do this count so
1:20:00
would usually have to do this count so that becomes a problem
1:20:01
that becomes a problem
1:20:01
that becomes a problem and so you look at you know what's
1:20:03
and so you look at you know what's
1:20:03
and so you look at you know what's available out there in terms of
1:20:04
available out there in terms of
1:20:04
available out there in terms of technology that can actually address
1:20:06
technology that can actually address
1:20:06
technology that can actually address this
1:20:07
this
1:20:07
this and so from that perspective you know
1:20:09
and so from that perspective you know
1:20:09
and so from that perspective you know you have to look at a
1:20:10
you have to look at a
1:20:10
you have to look at a machine learning opportunity or to
1:20:13
machine learning opportunity or to
1:20:13
machine learning opportunity or to understand
1:20:14
understand
1:20:14
understand how am i going to recognize all those
1:20:15
how am i going to recognize all those
1:20:15
how am i going to recognize all those objects and do these counts
1:20:17
objects and do these counts
1:20:17
objects and do these counts now i want to call um in perfect point a
1:20:20
now i want to call um in perfect point a
1:20:20
now i want to call um in perfect point a segway in regards to this uh call out to
1:20:22
segway in regards to this uh call out to
1:20:22
segway in regards to this uh call out to one of my colleagues uh casey um because
1:20:25
one of my colleagues uh casey um because
1:20:25
one of my colleagues uh casey um because she heard you know this story
1:20:26
she heard you know this story
1:20:26
she heard you know this story in terms of you know what i've done with
1:20:29
in terms of you know what i've done with
1:20:29
in terms of you know what i've done with the recognizing
1:20:30
the recognizing
1:20:30
the recognizing recognize recognization of objects uh to
1:20:33
recognize recognization of objects uh to
1:20:33
recognize recognization of objects uh to for inventory accounting and said wow
1:20:34
for inventory accounting and said wow
1:20:34
for inventory accounting and said wow you're you're going out to
1:20:36
you're you're going out to
1:20:36
you're you're going out to uh bing image search and you're pulling
1:20:37
uh bing image search and you're pulling
1:20:37
uh bing image search and you're pulling images one by one and creating this
1:20:40
images one by one and creating this
1:20:40
images one by one and creating this uh uh repository and then in putting it
1:20:42
uh uh repository and then in putting it
1:20:42
uh uh repository and then in putting it into the custom vision
1:20:44
into the custom vision
1:20:44
into the custom vision workbench for identification how about
1:20:46
workbench for identification how about
1:20:46
workbench for identification how about if i automate that for you and she wrote
1:20:48
if i automate that for you and she wrote
1:20:48
if i automate that for you and she wrote this great
1:20:49
this great
1:20:49
this great um blog post on dev.t.o uh that
1:20:52
um blog post on dev.t.o uh that
1:20:52
um blog post on dev.t.o uh that highlights you know the ability to
1:20:54
highlights you know the ability to
1:20:54
highlights you know the ability to extract information from uh from bing in
1:20:57
extract information from uh from bing in
1:20:57
extract information from uh from bing in terms of images
1:20:58
terms of images
1:20:58
terms of images and then have that made as a a
1:21:02
and then have that made as a a
1:21:02
and then have that made as a a repository for the learning of uh
1:21:05
repository for the learning of uh
1:21:05
repository for the learning of uh customvision.ai
1:21:06
customvision.ai
1:21:06
customvision.ai to identify objects and i just want to
1:21:08
to identify objects and i just want to
1:21:08
to identify objects and i just want to call it her brilliant work and again
1:21:09
call it her brilliant work and again
1:21:09
call it her brilliant work and again this was based on a need in terms of
1:21:12
this was based on a need in terms of
1:21:12
this was based on a need in terms of addressing an opportunity
1:21:14
addressing an opportunity
1:21:14
addressing an opportunity that existed that i was doing things
1:21:15
that existed that i was doing things
1:21:15
that existed that i was doing things manually uh as opposed to
1:21:17
manually uh as opposed to
1:21:17
manually uh as opposed to leading with technology and saying this
1:21:19
leading with technology and saying this
1:21:19
leading with technology and saying this is going to be the next greatest latest
1:21:20
is going to be the next greatest latest
1:21:20
is going to be the next greatest latest and greatest thing
1:21:21
and greatest thing
1:21:21
and greatest thing go and do this right so uh this you know
1:21:23
go and do this right so uh this you know
1:21:23
go and do this right so uh this you know this this this blog post that she has is
1:21:25
this this this blog post that she has is
1:21:25
this this this blog post that she has is awesome because it takes you through the
1:21:26
awesome because it takes you through the
1:21:26
awesome because it takes you through the steps in terms of
1:21:27
steps in terms of
1:21:27
steps in terms of this enablement and even somebody who's
1:21:29
this enablement and even somebody who's
1:21:29
this enablement and even somebody who's like me who just dabbles in code i i
1:21:31
like me who just dabbles in code i i
1:21:31
like me who just dabbles in code i i would say i'm a
1:21:33
would say i'm a
1:21:33
would say i'm a a hobbyist programmer at best uh has
1:21:35
a hobbyist programmer at best uh has
1:21:35
a hobbyist programmer at best uh has gone through this this scenario has been
1:21:37
gone through this this scenario has been
1:21:37
gone through this this scenario has been such a huge
1:21:38
such a huge
1:21:38
such a huge uh weight off my shoulders in terms of
1:21:39
uh weight off my shoulders in terms of
1:21:39
uh weight off my shoulders in terms of the acquisition of these images to teach
1:21:42
the acquisition of these images to teach
1:21:42
the acquisition of these images to teach uh machine learning what you know how to
1:21:43
uh machine learning what you know how to
1:21:43
uh machine learning what you know how to identify lab uh objects
1:21:45
identify lab uh objects
1:21:45
identify lab uh objects through custom vision ai so in terms of
1:21:48
through custom vision ai so in terms of
1:21:48
through custom vision ai so in terms of the demo itself i'll show you what it
1:21:50
the demo itself i'll show you what it
1:21:50
the demo itself i'll show you what it looks like so this is in essence the
1:21:51
looks like so this is in essence the
1:21:51
looks like so this is in essence the workbench that i would utilize or i
1:21:54
workbench that i would utilize or i
1:21:54
workbench that i would utilize or i would work with
1:21:55
would work with
1:21:55
would work with uh in regards to the identification of
1:21:57
uh in regards to the identification of
1:21:57
uh in regards to the identification of these tools or objects
1:21:58
these tools or objects
1:21:58
these tools or objects the same workbench was actually used in
1:22:00
the same workbench was actually used in
1:22:00
the same workbench was actually used in a more rudimentary form because it was
1:22:02
a more rudimentary form because it was
1:22:02
a more rudimentary form because it was four years ago
1:22:03
four years ago
1:22:03
four years ago uh in terms of the the identification of
1:22:05
uh in terms of the the identification of
1:22:05
uh in terms of the the identification of the life jacket as well
1:22:06
the life jacket as well
1:22:06
the life jacket as well and so in this scenario here i'm gonna
1:22:09
and so in this scenario here i'm gonna
1:22:09
and so in this scenario here i'm gonna go forth i've gone
1:22:10
go forth i've gone
1:22:10
go forth i've gone uh and used the solution that casey has
1:22:13
uh and used the solution that casey has
1:22:13
uh and used the solution that casey has built uh in regards to
1:22:14
built uh in regards to
1:22:14
built uh in regards to the collection of the images and i've
1:22:17
the collection of the images and i've
1:22:17
the collection of the images and i've gone forth and i grabbed the images and
1:22:18
gone forth and i grabbed the images and
1:22:18
gone forth and i grabbed the images and this you know
1:22:19
this you know
1:22:19
this you know for everybody at home and you know
1:22:21
for everybody at home and you know
1:22:21
for everybody at home and you know everybody in studio what what what are
1:22:22
everybody in studio what what what are
1:22:22
everybody in studio what what what are we looking at right now
1:22:24
we looking at right now
1:22:24
we looking at right now oh a bunch of hammers we're looking at a
1:22:26
oh a bunch of hammers we're looking at a
1:22:26
oh a bunch of hammers we're looking at a bunch of hammers right and how do you
1:22:27
bunch of hammers right and how do you
1:22:28
bunch of hammers right and how do you know it's a hammer
1:22:29
know it's a hammer
1:22:29
know it's a hammer uh it's usually the shape of the head
1:22:31
uh it's usually the shape of the head
1:22:31
uh it's usually the shape of the head and then and then there's a wooden
1:22:32
and then and then there's a wooden
1:22:32
and then and then there's a wooden handle or maybe a metal handle
1:22:34
handle or maybe a metal handle
1:22:34
handle or maybe a metal handle but it's sort of the shape of the head i
1:22:36
but it's sort of the shape of the head i
1:22:36
but it's sort of the shape of the head i guess that gives it gives it away
1:22:39
guess that gives it gives it away
1:22:39
guess that gives it gives it away right so you have a blunt object that's
1:22:41
right so you have a blunt object that's
1:22:41
right so you have a blunt object that's at the top right
1:22:42
at the top right
1:22:42
at the top right followed by a long handle uh yes there
1:22:45
followed by a long handle uh yes there
1:22:45
followed by a long handle uh yes there are different shapes of the hammer but
1:22:46
are different shapes of the hammer but
1:22:46
are different shapes of the hammer but that's in essence what we know as a
1:22:47
that's in essence what we know as a
1:22:47
that's in essence what we know as a hammer right we're taught that as young
1:22:49
hammer right we're taught that as young
1:22:49
hammer right we're taught that as young children
1:22:50
children
1:22:50
children to teach customvision.ai you actually
1:22:54
to teach customvision.ai you actually
1:22:54
to teach customvision.ai you actually have to go forth and grab all these
1:22:55
have to go forth and grab all these
1:22:55
have to go forth and grab all these images and then what we do is we
1:22:56
images and then what we do is we
1:22:56
images and then what we do is we actually label
1:22:57
actually label
1:22:57
actually label the images as what we know they are so
1:23:01
the images as what we know they are so
1:23:01
the images as what we know they are so using uh casey solution to pull that
1:23:04
using uh casey solution to pull that
1:23:04
using uh casey solution to pull that imagery out of bing
1:23:05
imagery out of bing
1:23:05
imagery out of bing uh to create this repository that i
1:23:07
uh to create this repository that i
1:23:07
uh to create this repository that i would then upload into
1:23:09
would then upload into
1:23:09
would then upload into my my custom vision.ai workbench
1:23:12
my my custom vision.ai workbench
1:23:12
my my custom vision.ai workbench label all these images as a hammer
1:23:15
label all these images as a hammer
1:23:15
label all these images as a hammer and then i'm going to add another object
1:23:18
and then i'm going to add another object
1:23:18
and then i'm going to add another object so what's what is the next object that
1:23:20
so what's what is the next object that
1:23:20
so what's what is the next object that we've just added uh
1:23:22
we've just added uh
1:23:22
we've just added uh a classic uh wrench so we've added a
1:23:25
a classic uh wrench so we've added a
1:23:25
a classic uh wrench so we've added a classic wrench to the to the mix now too
1:23:27
classic wrench to the to the mix now too
1:23:27
classic wrench to the to the mix now too and we've labeled you know the wrenches
1:23:30
and we've labeled you know the wrenches
1:23:30
and we've labeled you know the wrenches so that again custom vision ai gets an
1:23:31
so that again custom vision ai gets an
1:23:31
so that again custom vision ai gets an understanding of
1:23:32
understanding of
1:23:32
understanding of this is a hammer this is a wrench and
1:23:34
this is a hammer this is a wrench and
1:23:34
this is a hammer this is a wrench and we're doing this on behalf of tailwind
1:23:36
we're doing this on behalf of tailwind
1:23:36
we're doing this on behalf of tailwind traders
1:23:36
traders
1:23:36
traders to do that ability of understanding what
1:23:39
to do that ability of understanding what
1:23:39
to do that ability of understanding what i'm looking at so that it actually can
1:23:40
i'm looking at so that it actually can
1:23:40
i'm looking at so that it actually can go forth and do camps
1:23:42
go forth and do camps
1:23:42
go forth and do camps right so now we've taken these two
1:23:43
right so now we've taken these two
1:23:43
right so now we've taken these two objects and we're only going to do two
1:23:45
objects and we're only going to do two
1:23:45
objects and we're only going to do two for now because we can do you know a
1:23:46
for now because we can do you know a
1:23:46
for now because we can do you know a plethora of objects
1:23:47
plethora of objects
1:23:47
plethora of objects made available to us uh in terms of
1:23:49
made available to us uh in terms of
1:23:49
made available to us uh in terms of custom vision uh dot ai
1:23:51
custom vision uh dot ai
1:23:51
custom vision uh dot ai and how many power works of course
1:23:52
and how many power works of course
1:23:52
and how many power works of course limited by environment sorry
1:23:54
limited by environment sorry
1:23:54
limited by environment sorry uh so how many objects can you typically
1:23:56
uh so how many objects can you typically
1:23:56
uh so how many objects can you typically add to a data set in custom vision ai
1:23:58
add to a data set in custom vision ai
1:23:58
add to a data set in custom vision ai before it breaks down
1:24:01
before it breaks down
1:24:01
before it breaks down it's dependent on how it's going to be
1:24:03
it's dependent on how it's going to be
1:24:03
it's dependent on how it's going to be used
1:24:04
used
1:24:04
used right so in this scenario here what
1:24:07
right so in this scenario here what
1:24:07
right so in this scenario here what we've done in this specific
1:24:09
we've done in this specific
1:24:09
we've done in this specific uh project we were able to do 72 objects
1:24:13
uh project we were able to do 72 objects
1:24:13
uh project we were able to do 72 objects uh and i'll show you what the end result
1:24:14
uh and i'll show you what the end result
1:24:14
uh and i'll show you what the end result of this project actually is
1:24:16
of this project actually is
1:24:16
of this project actually is but the limitation is is based on your
1:24:19
but the limitation is is based on your
1:24:19
but the limitation is is based on your uh device and environment that you're
1:24:21
uh device and environment that you're
1:24:21
uh device and environment that you're going to be running this on
1:24:22
going to be running this on
1:24:22
going to be running this on we wanted to emulate this as close as
1:24:24
we wanted to emulate this as close as
1:24:24
we wanted to emulate this as close as possible to what was available on the
1:24:26
possible to what was available on the
1:24:26
possible to what was available on the drone
1:24:27
drone
1:24:27
drone uh in respect to the drone the drone
1:24:28
uh in respect to the drone the drone
1:24:28
uh in respect to the drone the drone only had to understand one object which
1:24:30
only had to understand one object which
1:24:30
only had to understand one object which was the life jacket that was the trigger
1:24:32
was the life jacket that was the trigger
1:24:32
was the life jacket that was the trigger and you know we had we had uh 5 000
1:24:35
and you know we had we had uh 5 000
1:24:35
and you know we had we had uh 5 000 worth of learning imagery for the drone
1:24:37
worth of learning imagery for the drone
1:24:37
worth of learning imagery for the drone in terms of different
1:24:38
in terms of different
1:24:38
in terms of different um environments cloudy day rainy days
1:24:41
um environments cloudy day rainy days
1:24:41
um environments cloudy day rainy days sunny day what have you
1:24:42
sunny day what have you
1:24:42
sunny day what have you uh but it was always just the life
1:24:43
uh but it was always just the life
1:24:43
uh but it was always just the life jacket right so the
1:24:45
jacket right so the
1:24:45
jacket right so the the learnings for the solution in terms
1:24:48
the learnings for the solution in terms
1:24:48
the learnings for the solution in terms of custom vision.ai
1:24:49
of custom vision.ai
1:24:49
of custom vision.ai was the whole aspect of learning to
1:24:51
was the whole aspect of learning to
1:24:51
was the whole aspect of learning to identify the one object
1:24:53
identify the one object
1:24:53
identify the one object in numerous environments in this
1:24:55
in numerous environments in this
1:24:55
in numerous environments in this scenario here it's on the flip side
1:24:57
scenario here it's on the flip side
1:24:57
scenario here it's on the flip side we're going to learn numerous
1:24:58
we're going to learn numerous
1:24:58
we're going to learn numerous um sorry numerous objects
1:25:02
um sorry numerous objects
1:25:02
um sorry numerous objects uh in a single environment so that you'd
1:25:05
uh in a single environment so that you'd
1:25:05
uh in a single environment so that you'd have to have proper light
1:25:06
have to have proper light
1:25:06
have to have proper light um so it was the flip side so now we
1:25:08
um so it was the flip side so now we
1:25:08
um so it was the flip side so now we have you know 72 objects that we're
1:25:10
have you know 72 objects that we're
1:25:10
have you know 72 objects that we're trying to
1:25:11
trying to
1:25:11
trying to recognize in a bright sunny you know lit
1:25:14
recognize in a bright sunny you know lit
1:25:14
recognize in a bright sunny you know lit up environment right so it's
1:25:16
up environment right so it's
1:25:16
up environment right so it's totally dependent on how the solution is
1:25:19
totally dependent on how the solution is
1:25:19
totally dependent on how the solution is going to come to play in terms of your
1:25:20
going to come to play in terms of your
1:25:20
going to come to play in terms of your your uh deployment of it that's why
1:25:23
your uh deployment of it that's why
1:25:23
your uh deployment of it that's why understanding the solution first as
1:25:25
understanding the solution first as
1:25:25
understanding the solution first as opposed to
1:25:26
opposed to
1:25:26
opposed to leading with the technology is so
1:25:27
leading with the technology is so
1:25:27
leading with the technology is so important because
1:25:29
important because
1:25:29
important because if you led with the technology and you
1:25:30
if you led with the technology and you
1:25:30
if you led with the technology and you come across this limitation then you're
1:25:32
come across this limitation then you're
1:25:32
come across this limitation then you're stuck
1:25:33
stuck
1:25:33
stuck but if you have a complete understanding
1:25:34
but if you have a complete understanding
1:25:34
but if you have a complete understanding of what you have available to you in
1:25:36
of what you have available to you in
1:25:36
of what you have available to you in your environment
1:25:37
your environment
1:25:37
your environment then you have a better success of
1:25:39
then you have a better success of
1:25:39
then you have a better success of achieving what you're trying to achieve
1:25:40
achieving what you're trying to achieve
1:25:40
achieving what you're trying to achieve in your proof of concept
1:25:41
in your proof of concept
1:25:41
in your proof of concept and then quickly go into production so
1:25:43
and then quickly go into production so
1:25:43
and then quickly go into production so actually in this case the tool is not so
1:25:46
actually in this case the tool is not so
1:25:46
actually in this case the tool is not so so much advanced in its usage but the
1:25:49
so much advanced in its usage but the
1:25:49
so much advanced in its usage but the scenario that you're going to use it in
1:25:51
scenario that you're going to use it in
1:25:51
scenario that you're going to use it in might make it very advanced i mean if
1:25:53
might make it very advanced i mean if
1:25:53
might make it very advanced i mean if you're talking about a flying drone in
1:25:55
you're talking about a flying drone in
1:25:55
you're talking about a flying drone in bad weather
1:25:56
bad weather
1:25:56
bad weather lighting is not ideal uh it might get
1:25:59
lighting is not ideal uh it might get
1:25:59
lighting is not ideal uh it might get lopsided pictures because it's blown
1:26:01
lopsided pictures because it's blown
1:26:01
lopsided pictures because it's blown away by the wind those kind of things
1:26:02
away by the wind those kind of things
1:26:02
away by the wind those kind of things that makes
1:26:03
that makes
1:26:03
that makes that makes this very advanced actually
1:26:06
that makes this very advanced actually
1:26:06
that makes this very advanced actually that's correct and so that's why i
1:26:09
that's correct and so that's why i
1:26:09
that's correct and so that's why i always
1:26:09
always
1:26:09
always suggest to understand your opportunity
1:26:11
suggest to understand your opportunity
1:26:11
suggest to understand your opportunity first before as as opposed to leading by
1:26:13
first before as as opposed to leading by
1:26:13
first before as as opposed to leading by technology
1:26:14
technology
1:26:14
technology because if you lead with technology you
1:26:16
because if you lead with technology you
1:26:16
because if you lead with technology you you forget about the nuances of
1:26:18
you forget about the nuances of
1:26:18
you forget about the nuances of a windy day in the drone having problems
1:26:20
a windy day in the drone having problems
1:26:20
a windy day in the drone having problems stabilizing
1:26:22
stabilizing
1:26:22
stabilizing yeah yeah it makes sense yeah so let's
1:26:25
yeah yeah it makes sense yeah so let's
1:26:25
yeah yeah it makes sense yeah so let's continue on
1:26:26
continue on
1:26:26
continue on uh so we've now identified you know the
1:26:28
uh so we've now identified you know the
1:26:28
uh so we've now identified you know the hammer and the wrench and we're going to
1:26:29
hammer and the wrench and we're going to
1:26:29
hammer and the wrench and we're going to train the model we're going to do two
1:26:31
train the model we're going to do two
1:26:31
train the model we're going to do two objects here for the demo today
1:26:32
objects here for the demo today
1:26:32
objects here for the demo today uh and i go through iterations of
1:26:34
uh and i go through iterations of
1:26:34
uh and i go through iterations of training so this is now computer vision
1:26:36
training so this is now computer vision
1:26:36
training so this is now computer vision going through
1:26:37
going through
1:26:37
going through doing an analysis of the objects that
1:26:39
doing an analysis of the objects that
1:26:39
doing an analysis of the objects that have been inputted uh and labeled
1:26:40
have been inputted uh and labeled
1:26:40
have been inputted uh and labeled uh as a hammer and a wrench and to have
1:26:43
uh as a hammer and a wrench and to have
1:26:43
uh as a hammer and a wrench and to have the ability to then come
1:26:44
the ability to then come
1:26:44
the ability to then come come back with a this is of what you've
1:26:46
come back with a this is of what you've
1:26:46
come back with a this is of what you've fed to me in terms of data
1:26:49
fed to me in terms of data
1:26:49
fed to me in terms of data the realization and understanding of
1:26:50
the realization and understanding of
1:26:50
the realization and understanding of what i'm looking at from a custom vision
1:26:53
what i'm looking at from a custom vision
1:26:53
what i'm looking at from a custom vision uh
1:26:53
uh
1:26:53
uh dot ai capability this means that with
1:26:56
dot ai capability this means that with
1:26:56
dot ai capability this means that with you know
1:26:57
you know
1:26:57
you know the solution of the two objects with a
1:26:58
the solution of the two objects with a
1:26:58
the solution of the two objects with a hundred percent precision to understand
1:27:01
hundred percent precision to understand
1:27:01
hundred percent precision to understand what what up what they're actually
1:27:02
what what up what they're actually
1:27:02
what what up what they're actually looking at what the computer vision is
1:27:03
looking at what the computer vision is
1:27:03
looking at what the computer vision is actually looking at
1:27:04
actually looking at
1:27:04
actually looking at uh and they have a recall of 96 which is
1:27:07
uh and they have a recall of 96 which is
1:27:07
uh and they have a recall of 96 which is very high
1:27:08
very high
1:27:08
very high uh when you have more objects you have
1:27:10
uh when you have more objects you have
1:27:10
uh when you have more objects you have to provide more objects
1:27:12
to provide more objects
1:27:12
to provide more objects um to teach uh the solution to
1:27:14
um to teach uh the solution to
1:27:14
um to teach uh the solution to understand
1:27:15
understand
1:27:15
understand what it's looking at so you do keep that
1:27:17
what it's looking at so you do keep that
1:27:17
what it's looking at so you do keep that into effect
1:27:18
into effect
1:27:18
into effect uh i know with the drone solution we
1:27:19
uh i know with the drone solution we
1:27:20
uh i know with the drone solution we were hovering around the 70 mark
1:27:22
were hovering around the 70 mark
1:27:22
were hovering around the 70 mark uh which was enough in terms of
1:27:24
uh which was enough in terms of
1:27:24
uh which was enough in terms of regulatory that was required
1:27:25
regulatory that was required
1:27:25
regulatory that was required uh for search and rescue um but 96 is
1:27:29
uh for search and rescue um but 96 is
1:27:29
uh for search and rescue um but 96 is very high
1:27:30
very high
1:27:30
very high it's very uncommon especially uh if you
1:27:32
it's very uncommon especially uh if you
1:27:32
it's very uncommon especially uh if you have multiples of objects that you have
1:27:34
have multiples of objects that you have
1:27:34
have multiples of objects that you have to put in the more objects you need to
1:27:35
to put in the more objects you need to
1:27:35
to put in the more objects you need to recognize the more learning or teaching
1:27:37
recognize the more learning or teaching
1:27:37
recognize the more learning or teaching that you have to
1:27:38
that you have to
1:27:38
that you have to to provide to uh the custom vision model
1:27:41
to provide to uh the custom vision model
1:27:41
to provide to uh the custom vision model to understand what it's looking at
1:27:43
to understand what it's looking at
1:27:43
to understand what it's looking at uh and do know that you know if the
1:27:45
uh and do know that you know if the
1:27:46
uh and do know that you know if the model can't
1:27:46
model can't
1:27:46
model can't you know if i have a hammer and a wrench
1:27:48
you know if i have a hammer and a wrench
1:27:48
you know if i have a hammer and a wrench and i and i introduced a screwdriver and
1:27:49
and i and i introduced a screwdriver and
1:27:50
and i and i introduced a screwdriver and i haven't taught
1:27:50
i haven't taught
1:27:50
i haven't taught uh the custom custom vision model what a
1:27:53
uh the custom custom vision model what a
1:27:53
uh the custom custom vision model what a screwdriver is
1:27:54
screwdriver is
1:27:54
screwdriver is it's not going to be able to identify
1:27:55
it's not going to be able to identify
1:27:55
it's not going to be able to identify we're only teaching it what we're
1:27:56
we're only teaching it what we're
1:27:56
we're only teaching it what we're teaching it and that's what it's going
1:27:57
teaching it and that's what it's going
1:27:57
teaching it and that's what it's going to know
1:27:58
to know
1:27:58
to know it will try to do best guess to say
1:28:00
it will try to do best guess to say
1:28:00
it will try to do best guess to say maybe it looks like a hammer
1:28:02
maybe it looks like a hammer
1:28:02
maybe it looks like a hammer maybe it looks like a wrench but its
1:28:03
maybe it looks like a wrench but its
1:28:03
maybe it looks like a wrench but its recall on that one object will be a lot
1:28:05
recall on that one object will be a lot
1:28:06
recall on that one object will be a lot lower
1:28:06
lower
1:28:06
lower and so when we talk about precision and
1:28:08
and so when we talk about precision and
1:28:08
and so when we talk about precision and recall in this case
1:28:11
recall in this case
1:28:11
recall in this case what would you say is the most important
1:28:14
what would you say is the most important
1:28:14
what would you say is the most important metric to keep in mind when we talk
1:28:15
metric to keep in mind when we talk
1:28:16
metric to keep in mind when we talk about a search and rescue mission
1:28:18
about a search and rescue mission
1:28:18
about a search and rescue mission is it is it a recall thing or is it yeah
1:28:20
is it is it a recall thing or is it yeah
1:28:20
is it is it a recall thing or is it yeah it would be recall
1:28:21
it would be recall
1:28:21
it would be recall the reason why it's recall uh because in
1:28:24
the reason why it's recall uh because in
1:28:24
the reason why it's recall uh because in precision there is the possibility of
1:28:26
precision there is the possibility of
1:28:26
precision there is the possibility of false
1:28:26
false
1:28:26
false positives in a recall scenario you're
1:28:29
positives in a recall scenario you're
1:28:29
positives in a recall scenario you're actually
1:28:30
actually
1:28:30
actually you know you're recalling your knowledge
1:28:32
you know you're recalling your knowledge
1:28:32
you know you're recalling your knowledge on a specific object
1:28:33
on a specific object
1:28:34
on a specific object so that it's certain it's it being
1:28:35
so that it's certain it's it being
1:28:35
so that it's certain it's it being certain about what it's looking at and
1:28:37
certain about what it's looking at and
1:28:37
certain about what it's looking at and it's percentage and i'll show you that
1:28:38
it's percentage and i'll show you that
1:28:38
it's percentage and i'll show you that in the example
1:28:39
in the example
1:28:39
in the example is very important because you can
1:28:40
is very important because you can
1:28:40
is very important because you can actually set that threshold to say if
1:28:42
actually set that threshold to say if
1:28:42
actually set that threshold to say if your recall is under 70 on the object
1:28:45
your recall is under 70 on the object
1:28:45
your recall is under 70 on the object that you're looking at
1:28:46
that you're looking at
1:28:46
that you're looking at it may not be the object that you're
1:28:48
it may not be the object that you're
1:28:48
it may not be the object that you're looking for yeah so in this case
1:28:50
looking for yeah so in this case
1:28:50
looking for yeah so in this case it's more important to actually find the
1:28:53
it's more important to actually find the
1:28:53
it's more important to actually find the person in the life jacket um
1:28:56
person in the life jacket um
1:28:56
person in the life jacket um even though it could be a false positive
1:28:58
even though it could be a false positive
1:28:58
even though it could be a false positive and it's actually not a person in the
1:28:59
and it's actually not a person in the
1:28:59
and it's actually not a person in the live check
1:28:59
live check
1:29:00
live check we don't want to miss anybody uh in the
1:29:02
we don't want to miss anybody uh in the
1:29:02
we don't want to miss anybody uh in the case of search and rescue
1:29:04
case of search and rescue
1:29:04
case of search and rescue yeah people often overlook this people
1:29:06
yeah people often overlook this people
1:29:06
yeah people often overlook this people tend to
1:29:08
tend to
1:29:08
tend to especially in the beginning and i did
1:29:09
especially in the beginning and i did
1:29:10
especially in the beginning and i did the same when i started working on
1:29:11
the same when i started working on
1:29:11
the same when i started working on machine learning i started looking at
1:29:13
machine learning i started looking at
1:29:13
machine learning i started looking at accuracy actually of this model how
1:29:15
accuracy actually of this model how
1:29:15
accuracy actually of this model how accurate is it
1:29:16
accurate is it
1:29:16
accurate is it but that's not always a good choice um
1:29:20
but that's not always a good choice um
1:29:20
but that's not always a good choice um in a search and rescue it might be more
1:29:22
in a search and rescue it might be more
1:29:22
in a search and rescue it might be more important to find everybody in a live
1:29:24
important to find everybody in a live
1:29:24
important to find everybody in a live vest
1:29:24
vest
1:29:24
vest versus making sure that it is a live
1:29:27
versus making sure that it is a live
1:29:27
versus making sure that it is a live vest
1:29:27
vest
1:29:27
vest in the case of the of the wrench and the
1:29:29
in the case of the of the wrench and the
1:29:29
in the case of the of the wrench and the hammer i guess it's the other way around
1:29:31
hammer i guess it's the other way around
1:29:31
hammer i guess it's the other way around we want to look at precision i want to
1:29:32
we want to look at precision i want to
1:29:32
we want to look at precision i want to make sure that it's indeed a hammer
1:29:34
make sure that it's indeed a hammer
1:29:34
make sure that it's indeed a hammer in every case that i get a hammer in my
1:29:37
in every case that i get a hammer in my
1:29:37
in every case that i get a hammer in my uh
1:29:37
uh
1:29:37
uh camera view correct because you want to
1:29:39
camera view correct because you want to
1:29:39
camera view correct because you want to be precise in regards to
1:29:41
be precise in regards to
1:29:41
be precise in regards to understanding how many hammers you have
1:29:43
understanding how many hammers you have
1:29:43
understanding how many hammers you have in inventory right
1:29:44
in inventory right
1:29:44
in inventory right that's why again it's so important to
1:29:46
that's why again it's so important to
1:29:46
that's why again it's so important to understand the opportunity as a whole in
1:29:48
understand the opportunity as a whole in
1:29:48
understand the opportunity as a whole in terms of what you're trying to address
1:29:49
terms of what you're trying to address
1:29:50
terms of what you're trying to address as opposed to leading with technology
1:29:51
as opposed to leading with technology
1:29:51
as opposed to leading with technology because there are different needs for
1:29:53
because there are different needs for
1:29:53
because there are different needs for different opportunities
1:29:54
different opportunities
1:29:54
different opportunities and if you're following the technology
1:29:55
and if you're following the technology
1:29:55
and if you're following the technology first and not taking the needs into
1:29:56
first and not taking the needs into
1:29:56
first and not taking the needs into consideration
1:29:57
consideration
1:29:57
consideration your proof of concept is not going to be
1:29:59
your proof of concept is not going to be
1:29:59
your proof of concept is not going to be successful
1:30:02
okay so let's continue on so now we've
1:30:05
okay so let's continue on so now we've
1:30:05
okay so let's continue on so now we've gotten the data and we've gotten the
1:30:06
gotten the data and we've gotten the
1:30:06
gotten the data and we've gotten the recall data we've got the precision data
1:30:08
recall data we've got the precision data
1:30:08
recall data we've got the precision data and
1:30:08
and
1:30:08
and as mentioned the precision data is
1:30:10
as mentioned the precision data is
1:30:10
as mentioned the precision data is important in this scenario by
1:30:11
important in this scenario by
1:30:11
important in this scenario by identifying the devices
1:30:13
identifying the devices
1:30:13
identifying the devices or the uh tools so that i can do the
1:30:15
or the uh tools so that i can do the
1:30:15
or the uh tools so that i can do the specific count
1:30:16
specific count
1:30:16
specific count i'm actually going to export this
1:30:18
i'm actually going to export this
1:30:18
i'm actually going to export this learning and as you can see through the
1:30:21
learning and as you can see through the
1:30:21
learning and as you can see through the customvision.ai workbench
1:30:23
customvision.ai workbench
1:30:23
customvision.ai workbench i have a plethora of choices that i can
1:30:25
i have a plethora of choices that i can
1:30:25
i have a plethora of choices that i can actually choose
1:30:27
actually choose
1:30:27
actually choose on how to export this learning i wanted
1:30:30
on how to export this learning i wanted
1:30:30
on how to export this learning i wanted to do this as close as possible to what
1:30:31
to do this as close as possible to what
1:30:31
to do this as close as possible to what we did with the drones and the drones
1:30:33
we did with the drones and the drones
1:30:33
we did with the drones and the drones being three hours out at sea
1:30:34
being three hours out at sea
1:30:34
being three hours out at sea uh at 14.4 kilobits per second
1:30:36
uh at 14.4 kilobits per second
1:30:36
uh at 14.4 kilobits per second connectivity
1:30:37
connectivity
1:30:37
connectivity it wasn't enough to do you know a
1:30:39
it wasn't enough to do you know a
1:30:39
it wasn't enough to do you know a real-time connection
1:30:41
real-time connection
1:30:41
real-time connection uh for machine learning to have an
1:30:42
uh for machine learning to have an
1:30:42
uh for machine learning to have an understanding of what the drones were
1:30:43
understanding of what the drones were
1:30:43
understanding of what the drones were looking at
1:30:44
looking at
1:30:44
looking at so in this scenario here what we
1:30:46
so in this scenario here what we
1:30:46
so in this scenario here what we actually did was we exported
1:30:48
actually did was we exported
1:30:48
actually did was we exported the learnings through the the tool as an
1:30:51
the learnings through the the tool as an
1:30:51
the learnings through the the tool as an onyx file now if you've dabbled in
1:30:55
onyx file now if you've dabbled in
1:30:55
onyx file now if you've dabbled in custom vision before you know that onyx
1:30:57
custom vision before you know that onyx
1:30:58
custom vision before you know that onyx is a great open source
1:30:59
is a great open source
1:30:59
is a great open source solution that's made available uh it can
1:31:01
solution that's made available uh it can
1:31:01
solution that's made available uh it can be used in the plethora different
1:31:03
be used in the plethora different
1:31:03
be used in the plethora different solutions and environments and what have
1:31:04
solutions and environments and what have
1:31:04
solutions and environments and what have you it's the most malleable one that
1:31:06
you it's the most malleable one that
1:31:06
you it's the most malleable one that i've worked with thus far
1:31:08
i've worked with thus far
1:31:08
i've worked with thus far in regards to incorporating the solution
1:31:09
in regards to incorporating the solution
1:31:09
in regards to incorporating the solution into other functionality and other
1:31:11
into other functionality and other
1:31:11
into other functionality and other architecture
1:31:12
architecture
1:31:12
architecture but as you can see there's you know
1:31:13
but as you can see there's you know
1:31:13
but as you can see there's you know tensorflow you can run this in the
1:31:15
tensorflow you can run this in the
1:31:15
tensorflow you can run this in the container
1:31:15
container
1:31:16
container there's even a um the core ml offering
1:31:18
there's even a um the core ml offering
1:31:18
there's even a um the core ml offering that's available on ios
1:31:20
that's available on ios
1:31:20
that's available on ios so if you're you know again you're
1:31:21
so if you're you know again you're
1:31:21
so if you're you know again you're enabling students that you know have
1:31:23
enabling students that you know have
1:31:23
enabling students that you know have iphones
1:31:23
iphones
1:31:24
iphones you have that availability to build out
1:31:25
you have that availability to build out
1:31:25
you have that availability to build out a solution for them to
1:31:27
a solution for them to
1:31:27
a solution for them to you know adopt uh and and run with in in
1:31:30
you know adopt uh and and run with in in
1:31:30
you know adopt uh and and run with in in terms of classroom
1:31:31
terms of classroom
1:31:31
terms of classroom if that's you know the device of choice
1:31:33
if that's you know the device of choice
1:31:33
if that's you know the device of choice the beauty of going with onyx however is
1:31:35
the beauty of going with onyx however is
1:31:35
the beauty of going with onyx however is something that where it works on
1:31:36
something that where it works on
1:31:36
something that where it works on everything and anything and for me the
1:31:38
everything and anything and for me the
1:31:38
everything and anything and for me the recall time on onyx has been the
1:31:40
recall time on onyx has been the
1:31:40
recall time on onyx has been the quickest
1:31:40
quickest
1:31:40
quickest uh in terms of understanding what we're
1:31:42
uh in terms of understanding what we're
1:31:42
uh in terms of understanding what we're looking at with the solution as a whole
1:31:46
so in exporting it in this onyx.1.2 file
1:31:49
so in exporting it in this onyx.1.2 file
1:31:49
so in exporting it in this onyx.1.2 file we bring up our unity environment so
1:31:52
we bring up our unity environment so
1:31:52
we bring up our unity environment so again
1:31:53
again
1:31:53
again taking the same basis of the solution
1:31:55
taking the same basis of the solution
1:31:55
taking the same basis of the solution what we've done with the drone
1:31:56
what we've done with the drone
1:31:56
what we've done with the drone but instead of deploying it into a drone
1:31:59
but instead of deploying it into a drone
1:31:59
but instead of deploying it into a drone uh
1:31:59
uh
1:31:59
uh on on platform that's been available for
1:32:01
on on platform that's been available for
1:32:01
on on platform that's been available for the drone to understand what it's
1:32:03
the drone to understand what it's
1:32:03
the drone to understand what it's looking at
1:32:03
looking at
1:32:04
looking at we're going to take that same solution
1:32:05
we're going to take that same solution
1:32:05
we're going to take that same solution that same model file type capability
1:32:07
that same model file type capability
1:32:07
that same model file type capability and export it into a hall lens now a
1:32:10
and export it into a hall lens now a
1:32:10
and export it into a hall lens now a hololens as you know it's it's a windows
1:32:12
hololens as you know it's it's a windows
1:32:12
hololens as you know it's it's a windows 10 device it's it's got its own cpu its
1:32:14
10 device it's it's got its own cpu its
1:32:14
10 device it's it's got its own cpu its own operating system its own camera
1:32:16
own operating system its own camera
1:32:16
own operating system its own camera sense
1:32:16
sense
1:32:16
sense uh the big thing with the hololens as
1:32:17
uh the big thing with the hololens as
1:32:17
uh the big thing with the hololens as you know is the availability to interact
1:32:19
you know is the availability to interact
1:32:20
you know is the availability to interact uh with 3d models on the fly we call it
1:32:22
uh with 3d models on the fly we call it
1:32:22
uh with 3d models on the fly we call it augmented reality in terms of your
1:32:24
augmented reality in terms of your
1:32:24
augmented reality in terms of your environment
1:32:24
environment
1:32:24
environment you can walk through and interact with
1:32:26
you can walk through and interact with
1:32:26
you can walk through and interact with 3d models as if they were actually in
1:32:28
3d models as if they were actually in
1:32:28
3d models as if they were actually in front of you
1:32:29
front of you
1:32:30
front of you in this here this is again the gen 1
1:32:31
in this here this is again the gen 1
1:32:31
in this here this is again the gen 1 solution that we have available
1:32:33
solution that we have available
1:32:33
solution that we have available and i've taken uh the availability of my
1:32:36
and i've taken uh the availability of my
1:32:36
and i've taken uh the availability of my github repo
1:32:37
github repo
1:32:37
github repo uh through the community i've worked
1:32:39
uh through the community i've worked
1:32:39
uh through the community i've worked with
1:32:40
with
1:32:40
with uh they have you know a code that's been
1:32:42
uh they have you know a code that's been
1:32:42
uh they have you know a code that's been available and they were allowed
1:32:43
available and they were allowed
1:32:43
available and they were allowed but they allow it and that's the beauty
1:32:44
but they allow it and that's the beauty
1:32:44
but they allow it and that's the beauty of open source they allowed it for
1:32:46
of open source they allowed it for
1:32:46
of open source they allowed it for others to
1:32:46
others to
1:32:46
others to to um to dabble with so i've actually
1:32:49
to um to dabble with so i've actually
1:32:49
to um to dabble with so i've actually made available the original
1:32:50
made available the original
1:32:50
made available the original code that's available in terms of this
1:32:53
code that's available in terms of this
1:32:53
code that's available in terms of this solution
1:32:53
solution
1:32:54
solution which everything here works in terms of
1:32:55
which everything here works in terms of
1:32:55
which everything here works in terms of the model file and made available
1:32:57
the model file and made available
1:32:57
the model file and made available um but the output as you know is is
1:32:59
um but the output as you know is is
1:32:59
um but the output as you know is is going through a hall lens
1:33:00
going through a hall lens
1:33:00
going through a hall lens for the object recognition and i'll give
1:33:02
for the object recognition and i'll give
1:33:02
for the object recognition and i'll give you a quick example of what that looks
1:33:04
you a quick example of what that looks
1:33:04
you a quick example of what that looks like
1:33:04
like
1:33:04
like and remember we talked about the recall
1:33:06
and remember we talked about the recall
1:33:06
and remember we talked about the recall and the precision as you can see here
1:33:07
and the precision as you can see here
1:33:07
and the precision as you can see here we're gonna look at two objects we're
1:33:08
we're gonna look at two objects we're
1:33:08
we're gonna look at two objects we're gonna look at a soccer ball and a
1:33:09
gonna look at a soccer ball and a
1:33:09
gonna look at a soccer ball and a basketball
1:33:10
basketball
1:33:10
basketball and it's gonna do identification of each
1:33:12
and it's gonna do identification of each
1:33:12
and it's gonna do identification of each pay attention to the recall which is the
1:33:14
pay attention to the recall which is the
1:33:14
pay attention to the recall which is the percentages at the bottom there
1:33:15
percentages at the bottom there
1:33:15
percentages at the bottom there uh because this device was taught 72
1:33:18
uh because this device was taught 72
1:33:18
uh because this device was taught 72 like i said the project this
1:33:19
like i said the project this
1:33:19
like i said the project this proof of concept was you know 72 objects
1:33:21
proof of concept was you know 72 objects
1:33:22
proof of concept was you know 72 objects that it was going to go through and
1:33:22
that it was going to go through and
1:33:22
that it was going to go through and recognize
1:33:24
recognize
1:33:24
recognize and then going through it like i can
1:33:25
and then going through it like i can
1:33:25
and then going through it like i can recognize that this is a basketball
1:33:27
recognize that this is a basketball
1:33:27
recognize that this is a basketball 69 percent accuracy right so the recall
1:33:30
69 percent accuracy right so the recall
1:33:30
69 percent accuracy right so the recall is it it's a lot lower so i would say
1:33:33
is it it's a lot lower so i would say
1:33:33
is it it's a lot lower so i would say you know what it didn't do a good job at
1:33:34
you know what it didn't do a good job at
1:33:34
you know what it didn't do a good job at that
1:33:35
that
1:33:35
that this one it knows it's a soccer ball at
1:33:36
this one it knows it's a soccer ball at
1:33:36
this one it knows it's a soccer ball at 100 right i wanted to showcase that
1:33:39
100 right i wanted to showcase that
1:33:39
100 right i wanted to showcase that difference between the two
1:33:40
difference between the two
1:33:40
difference between the two because in the scenario of the life
1:33:42
because in the scenario of the life
1:33:42
because in the scenario of the life jacket you want to be you know
1:33:43
jacket you want to be you know
1:33:43
jacket you want to be you know in terms of your recall hey how many
1:33:46
in terms of your recall hey how many
1:33:46
in terms of your recall hey how many these life jackets do we have
1:33:47
these life jackets do we have
1:33:47
these life jackets do we have and then from the precision standpoint
1:33:49
and then from the precision standpoint
1:33:49
and then from the precision standpoint do we say okay we're going to drop the
1:33:50
do we say okay we're going to drop the
1:33:50
do we say okay we're going to drop the precision to 60
1:33:51
precision to 60
1:33:52
precision to 60 because lives matter we want to make
1:33:53
because lives matter we want to make
1:33:53
because lives matter we want to make sure that we we can find everybody
1:33:55
sure that we we can find everybody
1:33:55
sure that we we can find everybody that is you know in distress that we can
1:33:57
that is you know in distress that we can
1:33:57
that is you know in distress that we can rescue them immediately
1:33:58
rescue them immediately
1:33:58
rescue them immediately that was you know the beauty of the
1:34:00
that was you know the beauty of the
1:34:00
that was you know the beauty of the capability of
1:34:01
capability of
1:34:01
capability of doing the um threshold setting of
1:34:05
doing the um threshold setting of
1:34:05
doing the um threshold setting of precision and recall was of great
1:34:07
precision and recall was of great
1:34:07
precision and recall was of great importance especially in the drone
1:34:08
importance especially in the drone
1:34:08
importance especially in the drone scenario
1:34:09
scenario
1:34:09
scenario and of course in the in a in a scenario
1:34:10
and of course in the in a in a scenario
1:34:10
and of course in the in a in a scenario where we're doing inventory
1:34:12
where we're doing inventory
1:34:12
where we're doing inventory uh in regards to being able to identify
1:34:14
uh in regards to being able to identify
1:34:14
uh in regards to being able to identify objects to actually run account on them
1:34:19
objects to actually run account on them
1:34:19
objects to actually run account on them now oops a little too fast so i wanted
1:34:22
now oops a little too fast so i wanted
1:34:22
now oops a little too fast so i wanted to share this
1:34:22
to share this
1:34:22
to share this and this is the computer vision tool set
1:34:25
and this is the computer vision tool set
1:34:25
and this is the computer vision tool set that's made available
1:34:26
that's made available
1:34:26
that's made available uh if you're like me and you'd like to
1:34:28
uh if you're like me and you'd like to
1:34:28
uh if you're like me and you'd like to dabble uh
1:34:29
dabble uh
1:34:29
dabble uh in terms of hands-on and understanding
1:34:31
in terms of hands-on and understanding
1:34:31
in terms of hands-on and understanding the functionality and capability that's
1:34:32
the functionality and capability that's
1:34:32
the functionality and capability that's available out there
1:34:33
available out there
1:34:34
available out there um you know this is a great solution
1:34:35
um you know this is a great solution
1:34:35
um you know this is a great solution this is a curation of resources that are
1:34:37
this is a curation of resources that are
1:34:37
this is a curation of resources that are made available specifically to computer
1:34:39
made available specifically to computer
1:34:39
made available specifically to computer vision
1:34:39
vision
1:34:40
vision uh that you can go for the double dabble
1:34:41
uh that you can go for the double dabble
1:34:41
uh that you can go for the double dabble with also you know capture that
1:34:43
with also you know capture that
1:34:44
with also you know capture that github repo link that i provided and the
1:34:46
github repo link that i provided and the
1:34:46
github repo link that i provided and the drone link
1:34:47
drone link
1:34:47
drone link i provided tells a story and showcases
1:34:49
i provided tells a story and showcases
1:34:49
i provided tells a story and showcases the architecture
1:34:50
the architecture
1:34:50
the architecture of how the solution was built all of
1:34:52
of how the solution was built all of
1:34:52
of how the solution was built all of which you know
1:34:53
which you know
1:34:54
which you know go out and and replicate but go out and
1:34:57
go out and and replicate but go out and
1:34:57
go out and and replicate but go out and start
1:34:57
start
1:34:57
start imagining go out and start understanding
1:34:59
imagining go out and start understanding
1:34:59
imagining go out and start understanding you know hey these are a whole bunch of
1:35:01
you know hey these are a whole bunch of
1:35:01
you know hey these are a whole bunch of opportunities that i can address
1:35:03
opportunities that i can address
1:35:03
opportunities that i can address knowing what this technology can do
1:35:05
knowing what this technology can do
1:35:05
knowing what this technology can do again don't leave with the technology
1:35:07
again don't leave with the technology
1:35:07
again don't leave with the technology leave with the problem understanding of
1:35:09
leave with the problem understanding of
1:35:09
leave with the problem understanding of the problem or understanding of the
1:35:10
the problem or understanding of the
1:35:10
the problem or understanding of the opportunity first
1:35:11
opportunity first
1:35:11
opportunity first and then see where the technology can be
1:35:13
and then see where the technology can be
1:35:13
and then see where the technology can be can be fit
1:35:14
can be fit
1:35:14
can be fit and i challenge you know everyone that's
1:35:16
and i challenge you know everyone that's
1:35:16
and i challenge you know everyone that's here on on uh
1:35:17
here on on uh
1:35:18
here on on uh this this global global ai uh community
1:35:21
this this global global ai uh community
1:35:21
this this global global ai uh community go forth and teach me go forth and show
1:35:24
go forth and teach me go forth and show
1:35:24
go forth and teach me go forth and show me
1:35:25
me
1:35:25
me what you're trying to accomplish and you
1:35:26
what you're trying to accomplish and you
1:35:26
what you're trying to accomplish and you know i'm happy to provide resources to
1:35:28
know i'm happy to provide resources to
1:35:28
know i'm happy to provide resources to you in regards to
1:35:29
you in regards to
1:35:30
you in regards to upskilling or or um documentation or
1:35:32
upskilling or or um documentation or
1:35:32
upskilling or or um documentation or what have you
1:35:33
what have you
1:35:33
what have you uh in regards to trying to achieve what
1:35:35
uh in regards to trying to achieve what
1:35:36
uh in regards to trying to achieve what you're trying to achieve
1:35:37
you're trying to achieve
1:35:37
you're trying to achieve but you gotta let us know first you
1:35:38
but you gotta let us know first you
1:35:38
but you gotta let us know first you gotta let you know provide us that
1:35:40
gotta let you know provide us that
1:35:40
gotta let you know provide us that feedback
1:35:41
feedback
1:35:41
feedback so that we can understand how we can
1:35:43
so that we can understand how we can
1:35:43
so that we can understand how we can better you know help you
1:35:44
better you know help you
1:35:44
better you know help you in terms of services what needs to be
1:35:46
in terms of services what needs to be
1:35:46
in terms of services what needs to be modified what needs to be created what
1:35:47
modified what needs to be created what
1:35:47
modified what needs to be created what needs to be addressed
1:35:49
needs to be addressed
1:35:49
needs to be addressed uh so that you can achieve what you're
1:35:50
uh so that you can achieve what you're
1:35:50
uh so that you can achieve what you're trying to achieve
1:35:53
trying to achieve
1:35:53
trying to achieve thanks um i do have one last question
1:35:56
thanks um i do have one last question
1:35:56
thanks um i do have one last question and there and then we'll move on so
1:35:59
and there and then we'll move on so
1:35:59
and there and then we'll move on so um we've talked about detecting life
1:36:02
um we've talked about detecting life
1:36:02
um we've talked about detecting life vests
1:36:03
vests
1:36:03
vests um uh would it be especially
1:36:06
um uh would it be especially
1:36:06
um uh would it be especially hard to recognize somebody without a
1:36:07
hard to recognize somebody without a
1:36:07
hard to recognize somebody without a life vest i'm asking this because i'm
1:36:09
life vest i'm asking this because i'm
1:36:09
life vest i'm asking this because i'm wondering
1:36:10
wondering
1:36:10
wondering how easy is it for a a machine learning
1:36:13
how easy is it for a a machine learning
1:36:13
how easy is it for a a machine learning model like
1:36:14
model like
1:36:14
model like like the one that's used by computer
1:36:15
like the one that's used by computer
1:36:15
like the one that's used by computer custom computer vision to recognize
1:36:18
custom computer vision to recognize
1:36:18
custom computer vision to recognize um sort of difficult to see objects
1:36:22
um sort of difficult to see objects
1:36:22
um sort of difficult to see objects in a frame
1:36:25
in a frame
1:36:25
in a frame so the live jacket piece was an
1:36:27
so the live jacket piece was an
1:36:28
so the live jacket piece was an interesting one because
1:36:29
interesting one because
1:36:29
interesting one because we had the ability to identify a marker
1:36:31
we had the ability to identify a marker
1:36:31
we had the ability to identify a marker and the marker was the life jacket
1:36:33
and the marker was the life jacket
1:36:33
and the marker was the life jacket we also had to follow that up with an ir
1:36:35
we also had to follow that up with an ir
1:36:35
we also had to follow that up with an ir skin because identification of a life
1:36:37
skin because identification of a life
1:36:37
skin because identification of a life jacket wasn't enough because you can
1:36:38
jacket wasn't enough because you can
1:36:38
jacket wasn't enough because you can count four life jackets in the water
1:36:40
count four life jackets in the water
1:36:40
count four life jackets in the water but you may or may not have an
1:36:41
but you may or may not have an
1:36:41
but you may or may not have an individual even in that life jacket
1:36:42
individual even in that life jacket
1:36:42
individual even in that life jacket which is another problem right so
1:36:44
which is another problem right so
1:36:44
which is another problem right so i've sent out the resources for four
1:36:46
i've sent out the resources for four
1:36:46
i've sent out the resources for four life jackets that flew off the deck
1:36:47
life jackets that flew off the deck
1:36:47
life jackets that flew off the deck because it was
1:36:48
because it was
1:36:48
because it was it was a windy day uh and they're and
1:36:50
it was a windy day uh and they're and
1:36:50
it was a windy day uh and they're and you know the ship is in distress but
1:36:51
you know the ship is in distress but
1:36:51
you know the ship is in distress but it's because
1:36:52
it's because
1:36:52
it's because sorry the engine has stopped not because
1:36:54
sorry the engine has stopped not because
1:36:54
sorry the engine has stopped not because there were people in the water
1:36:56
there were people in the water
1:36:56
there were people in the water the trigger was a life jacket the added
1:36:59
the trigger was a life jacket the added
1:36:59
the trigger was a life jacket the added piece to that was an ir skin to detect
1:37:02
piece to that was an ir skin to detect
1:37:02
piece to that was an ir skin to detect the heat mass
1:37:03
the heat mass
1:37:03
the heat mass emits that life jacket to understand if
1:37:04
emits that life jacket to understand if
1:37:04
emits that life jacket to understand if it was an actual body or an individual
1:37:06
it was an actual body or an individual
1:37:06
it was an actual body or an individual in that life jacket and so yes you do
1:37:09
in that life jacket and so yes you do
1:37:09
in that life jacket and so yes you do have the capability to have
1:37:11
have the capability to have
1:37:11
have the capability to have an ir scan done but without the life
1:37:14
an ir scan done but without the life
1:37:14
an ir scan done but without the life jacket as a trigger
1:37:15
jacket as a trigger
1:37:15
jacket as a trigger that becomes then the challenge of the
1:37:17
that becomes then the challenge of the
1:37:17
that becomes then the challenge of the drone to survey the area to have a
1:37:19
drone to survey the area to have a
1:37:19
drone to survey the area to have a better understanding
1:37:20
better understanding
1:37:20
better understanding of what it's actually looking at right
1:37:22
of what it's actually looking at right
1:37:22
of what it's actually looking at right it can be done 100
1:37:24
it can be done 100
1:37:24
it can be done 100 but is your restriction time and how
1:37:27
but is your restriction time and how
1:37:27
but is your restriction time and how much
1:37:27
much
1:37:27
much much more time would it take for those
1:37:29
much more time would it take for those
1:37:29
much more time would it take for those drones to identify you know
1:37:31
drones to identify you know
1:37:31
drones to identify you know individuals in the water without life
1:37:32
individuals in the water without life
1:37:32
individuals in the water without life jackets that's something that you would
1:37:34
jackets that's something that you would
1:37:34
jackets that's something that you would have to take into consideration for for
1:37:35
have to take into consideration for for
1:37:35
have to take into consideration for for your project
1:37:37
your project
1:37:37
your project and and um from what i understand um
1:37:40
and and um from what i understand um
1:37:40
and and um from what i understand um is that instead of building one model to
1:37:43
is that instead of building one model to
1:37:43
is that instead of building one model to do
1:37:44
do
1:37:44
do everything you actually build a model
1:37:47
everything you actually build a model
1:37:47
everything you actually build a model you split the model in multiple parts so
1:37:49
you split the model in multiple parts so
1:37:49
you split the model in multiple parts so first you take a trigger
1:37:51
first you take a trigger
1:37:51
first you take a trigger that decides this is interesting or not
1:37:53
that decides this is interesting or not
1:37:53
that decides this is interesting or not for my drone
1:37:54
for my drone
1:37:54
for my drone and then you build a second model that's
1:37:56
and then you build a second model that's
1:37:56
and then you build a second model that's actually using infrared
1:37:57
actually using infrared
1:37:57
actually using infrared to make a decision is this a human or
1:37:59
to make a decision is this a human or
1:38:00
to make a decision is this a human or not so it's going from interesting
1:38:02
not so it's going from interesting
1:38:02
not so it's going from interesting object to
1:38:03
object to
1:38:03
object to a human in distress and that's kind of
1:38:04
a human in distress and that's kind of
1:38:04
a human in distress and that's kind of cool that you broke down this problem
1:38:06
cool that you broke down this problem
1:38:06
cool that you broke down this problem into smaller bits
1:38:08
into smaller bits
1:38:08
into smaller bits from my experience this is one of the
1:38:09
from my experience this is one of the
1:38:09
from my experience this is one of the key tricks that you have to apply to
1:38:11
key tricks that you have to apply to
1:38:11
key tricks that you have to apply to your ai
1:38:12
your ai
1:38:12
your ai break down your problem in smaller
1:38:14
break down your problem in smaller
1:38:14
break down your problem in smaller pieces that will help you
1:38:15
pieces that will help you
1:38:15
pieces that will help you a computer can only do so much that
1:38:18
a computer can only do so much that
1:38:18
a computer can only do so much that that's what i gather from
1:38:20
that's what i gather from
1:38:20
that's what i gather from you showing off the computer vision but
1:38:22
you showing off the computer vision but
1:38:22
you showing off the computer vision but also the stuff with the generating
1:38:23
also the stuff with the generating
1:38:23
also the stuff with the generating uh of objects in h3d space
1:38:27
uh of objects in h3d space
1:38:27
uh of objects in h3d space um yeah that's kind of cool and add to
1:38:30
um yeah that's kind of cool and add to
1:38:30
um yeah that's kind of cool and add to that complexity what we had available to
1:38:32
that complexity what we had available to
1:38:32
that complexity what we had available to us in terms of hardware
1:38:33
us in terms of hardware
1:38:33
us in terms of hardware so an arm processor uh with a two gig of
1:38:36
so an arm processor uh with a two gig of
1:38:36
so an arm processor uh with a two gig of ram
1:38:37
ram
1:38:37
ram and uh a was it again it was a 16 gb 16
1:38:40
and uh a was it again it was a 16 gb 16
1:38:40
and uh a was it again it was a 16 gb 16 gig micro sd for storage
1:38:42
gig micro sd for storage
1:38:42
gig micro sd for storage you know that's what we had to play with
1:38:44
you know that's what we had to play with
1:38:44
you know that's what we had to play with in terms of an environment to
1:38:46
in terms of an environment to
1:38:46
in terms of an environment to have the operating system run and and
1:38:49
have the operating system run and and
1:38:49
have the operating system run and and have the op
1:38:49
have the op
1:38:49
have the op the application understand the life
1:38:51
the application understand the life
1:38:51
the application understand the life jacket understand the ir
1:38:53
jacket understand the ir
1:38:53
jacket understand the ir do the calculation on the fly uh in
1:38:56
do the calculation on the fly uh in
1:38:56
do the calculation on the fly uh in terms of
1:38:57
terms of
1:38:57
terms of time till hypothermia sets in because it
1:38:59
time till hypothermia sets in because it
1:38:59
time till hypothermia sets in because it would capture environmentals as well
1:39:01
would capture environmentals as well
1:39:01
would capture environmentals as well and then report back immediately should
1:39:03
and then report back immediately should
1:39:03
and then report back immediately should it be
1:39:04
it be
1:39:04
it be a a critical time for those individuals
1:39:06
a a critical time for those individuals
1:39:06
a a critical time for those individuals that need to be rescued
1:39:09
that need to be rescued
1:39:09
that need to be rescued a very nice solution i'm so happy to to
1:39:12
a very nice solution i'm so happy to to
1:39:12
a very nice solution i'm so happy to to have you on the show so
1:39:13
have you on the show so
1:39:13
have you on the show so thank you very much uh there's a ton
1:39:16
thank you very much uh there's a ton
1:39:16
thank you very much uh there's a ton more questions that i have and i'm sure
1:39:17
more questions that i have and i'm sure
1:39:17
more questions that i have and i'm sure that people in the stream
1:39:19
that people in the stream
1:39:19
that people in the stream in the live chat also have more
1:39:21
in the live chat also have more
1:39:21
in the live chat also have more questions for you so feel free to stick
1:39:23
questions for you so feel free to stick
1:39:23
questions for you so feel free to stick around
1:39:23
around
1:39:23
around and we'll move on to our next
1:39:27
and we'll move on to our next
1:39:27
and we'll move on to our next guest uh noel silver
1:39:33
hey there hey hey i wasn't sure i'm like
1:39:37
hey there hey hey i wasn't sure i'm like
1:39:37
hey there hey hey i wasn't sure i'm like do i just jump on in
1:39:40
do i just jump on in
1:39:40
do i just jump on in was awesome so that was phenomenal like
1:39:42
was awesome so that was phenomenal like
1:39:42
was awesome so that was phenomenal like this is so cool
1:39:43
this is so cool
1:39:44
this is so cool so anyway thank you you're welcome
1:39:47
so anyway thank you you're welcome
1:39:47
so anyway thank you you're welcome welcome to the show go ahead
1:39:51
welcome to the show go ahead
1:39:51
welcome to the show go ahead oh i think uh he was just saying welcome
1:39:54
oh i think uh he was just saying welcome
1:39:54
oh i think uh he was just saying welcome so thank you for having me
1:39:55
so thank you for having me
1:39:55
so thank you for having me i'm super excited to be here cool
1:39:58
i'm super excited to be here cool
1:39:58
i'm super excited to be here cool we're we're delighted to welcome noelle
1:40:00
we're we're delighted to welcome noelle
1:40:00
we're we're delighted to welcome noelle silver and noelle silver
1:40:03
silver and noelle silver
1:40:03
silver and noelle silver has a long history and a long list of
1:40:05
has a long history and a long list of
1:40:05
has a long history and a long list of contributions
1:40:06
contributions
1:40:06
contributions to the ai community and many
1:40:10
to the ai community and many
1:40:10
to the ai community and many accolades with alexa skills and i know
1:40:14
accolades with alexa skills and i know
1:40:14
accolades with alexa skills and i know you have a heavy background in voice
1:40:16
you have a heavy background in voice
1:40:16
you have a heavy background in voice but i'm most excited about your
1:40:18
but i'm most excited about your
1:40:18
but i'm most excited about your contributions to women in tech and
1:40:19
contributions to women in tech and
1:40:20
contributions to women in tech and technology
1:40:21
technology
1:40:21
technology um i i love working in the dnai space
1:40:24
um i i love working in the dnai space
1:40:24
um i i love working in the dnai space because there there is just so many
1:40:26
because there there is just so many
1:40:26
because there there is just so many women in this space
1:40:27
women in this space
1:40:27
women in this space and i think that there's so many women
1:40:29
and i think that there's so many women
1:40:30
and i think that there's so many women because of women like you who make it a
1:40:31
because of women like you who make it a
1:40:31
because of women like you who make it a welcoming space so thank you noelle
1:40:34
welcoming space so thank you noelle
1:40:34
welcoming space so thank you noelle thank you that's awesome well yeah i
1:40:36
thank you that's awesome well yeah i
1:40:36
thank you that's awesome well yeah i hope there's more of us i always say
1:40:38
hope there's more of us i always say
1:40:38
hope there's more of us i always say like come on
1:40:39
like come on
1:40:39
like come on especially now anthony did a great job
1:40:41
especially now anthony did a great job
1:40:41
especially now anthony did a great job of explaining like
1:40:42
of explaining like
1:40:42
of explaining like you don't have to be any special person
1:40:45
you don't have to be any special person
1:40:45
you don't have to be any special person like you just have to know a problem or
1:40:46
like you just have to know a problem or
1:40:46
like you just have to know a problem or be empathetic to a problem to be able to
1:40:48
be empathetic to a problem to be able to
1:40:48
be empathetic to a problem to be able to solve it with these technologies and
1:40:50
solve it with these technologies and
1:40:50
solve it with these technologies and so i'm just like open the gates let them
1:40:53
so i'm just like open the gates let them
1:40:53
so i'm just like open the gates let them in
1:40:57
yeah so we've seen that ethics has
1:41:01
yeah so we've seen that ethics has
1:41:01
yeah so we've seen that ethics has been a big consideration when designing
1:41:05
been a big consideration when designing
1:41:05
been a big consideration when designing these ai systems
1:41:06
these ai systems
1:41:06
these ai systems and um we're looking forward to your
1:41:08
and um we're looking forward to your
1:41:08
and um we're looking forward to your session on explainable ai
1:41:11
session on explainable ai
1:41:11
session on explainable ai absolutely do i just like go into it and
1:41:15
absolutely do i just like go into it and
1:41:15
absolutely do i just like go into it and show it or what should i do
1:41:16
show it or what should i do
1:41:16
show it or what should i do absolutely um i don't know what you
1:41:19
absolutely um i don't know what you
1:41:19
absolutely um i don't know what you brought did you bring a set of slides
1:41:20
brought did you bring a set of slides
1:41:20
brought did you bring a set of slides first can i share my screen
1:41:22
first can i share my screen
1:41:22
first can i share my screen yeah go ahead all right i'm gonna do it
1:41:26
yeah go ahead all right i'm gonna do it
1:41:26
yeah go ahead all right i'm gonna do it let's see let's hope it works out um
1:41:28
let's see let's hope it works out um
1:41:28
let's see let's hope it works out um there'll be a little
1:41:30
there'll be a little
1:41:30
there'll be a little uh oh no there we go i think you could
1:41:32
uh oh no there we go i think you could
1:41:32
uh oh no there we go i think you could see it can you see it
1:41:33
see it can you see it
1:41:33
see it can you see it yes absolutely that's me okay
1:41:37
yes absolutely that's me okay
1:41:37
yes absolutely that's me okay and my people so um i'll just get
1:41:39
and my people so um i'll just get
1:41:39
and my people so um i'll just get started i am here to talk very
1:41:41
started i am here to talk very
1:41:42
started i am here to talk very we don't have a whole lot of time um and
1:41:43
we don't have a whole lot of time um and
1:41:44
we don't have a whole lot of time um and so i want to talk
1:41:45
so i want to talk
1:41:45
so i want to talk about a responsible ai in general but
1:41:48
about a responsible ai in general but
1:41:48
about a responsible ai in general but very specifically
1:41:49
very specifically
1:41:49
very specifically about what data scientists and modelers
1:41:51
about what data scientists and modelers
1:41:51
about what data scientists and modelers can do
1:41:52
can do
1:41:52
can do to build more explainable models but
1:41:54
to build more explainable models but
1:41:54
to build more explainable models but before that i always like to give a very
1:41:56
before that i always like to give a very
1:41:56
before that i always like to give a very brief you know
1:41:57
brief you know
1:41:57
brief you know two minutes of who i am and why i even
1:42:00
two minutes of who i am and why i even
1:42:00
two minutes of who i am and why i even got into this space so thank you alicia
1:42:02
got into this space so thank you alicia
1:42:02
got into this space so thank you alicia for telling
1:42:03
for telling
1:42:03
for telling everyone about like my start which was
1:42:06
everyone about like my start which was
1:42:06
everyone about like my start which was in alexa i was an early member of the
1:42:08
in alexa i was an early member of the
1:42:08
in alexa i was an early member of the alexa team and part of being
1:42:12
alexa team and part of being
1:42:12
alexa team and part of being part of that organization we didn't know
1:42:14
part of that organization we didn't know
1:42:14
part of that organization we didn't know what we were you know getting ourselves
1:42:16
what we were you know getting ourselves
1:42:16
what we were you know getting ourselves into
1:42:16
into
1:42:16
into the product hadn't been proven yet it
1:42:18
the product hadn't been proven yet it
1:42:18
the product hadn't been proven yet it was still in beta
1:42:19
was still in beta
1:42:19
was still in beta we had less than a thousand users like
1:42:21
we had less than a thousand users like
1:42:21
we had less than a thousand users like it was a baby product and
1:42:24
it was a baby product and
1:42:24
it was a baby product and we as i was on the data science team for
1:42:27
we as i was on the data science team for
1:42:27
we as i was on the data science team for a couple of years
1:42:28
a couple of years
1:42:28
a couple of years and we were just trying to like keep up
1:42:31
and we were just trying to like keep up
1:42:31
and we were just trying to like keep up and
1:42:31
and
1:42:31
and we weren't really having a lot of luxury
1:42:34
we weren't really having a lot of luxury
1:42:34
we weren't really having a lot of luxury of being able to
1:42:35
of being able to
1:42:35
of being able to sit back take time and think about how
1:42:38
sit back take time and think about how
1:42:38
sit back take time and think about how we were going to do everything
1:42:39
we were going to do everything
1:42:40
we were going to do everything every little thing but so now i'm pretty
1:42:42
every little thing but so now i'm pretty
1:42:42
every little thing but so now i'm pretty passionate about sharing with people
1:42:43
passionate about sharing with people
1:42:43
passionate about sharing with people here's what you could do if you find
1:42:45
here's what you could do if you find
1:42:45
here's what you could do if you find yourself in a situation like this
1:42:47
yourself in a situation like this
1:42:47
yourself in a situation like this and that's what i'll focus on over the
1:42:48
and that's what i'll focus on over the
1:42:48
and that's what i'll focus on over the next 20 minutes or so
1:42:50
next 20 minutes or so
1:42:50
next 20 minutes or so um but before that uh the reason i even
1:42:53
um but before that uh the reason i even
1:42:53
um but before that uh the reason i even got to alexa
1:42:54
got to alexa
1:42:54
got to alexa is because i had two people in my life
1:42:56
is because i had two people in my life
1:42:56
is because i had two people in my life that needed
1:42:58
that needed
1:42:58
that needed solutions very similar to what anthony
1:43:00
solutions very similar to what anthony
1:43:00
solutions very similar to what anthony was talking about like
1:43:01
was talking about like
1:43:01
was talking about like there was a problem that needed to be
1:43:02
there was a problem that needed to be
1:43:02
there was a problem that needed to be solved my son
1:43:04
solved my son
1:43:04
solved my son max he's there with the heart um he has
1:43:06
max he's there with the heart um he has
1:43:06
max he's there with the heart um he has down syndrome and he was really
1:43:08
down syndrome and he was really
1:43:08
down syndrome and he was really struggling as he got he's now
1:43:10
struggling as he got he's now
1:43:10
struggling as he got he's now 15 years old as he starts to get into
1:43:13
15 years old as he starts to get into
1:43:13
15 years old as he starts to get into kind of high school education curriculum
1:43:14
kind of high school education curriculum
1:43:14
kind of high school education curriculum and he's brilliant he's very smart
1:43:16
and he's brilliant he's very smart
1:43:16
and he's brilliant he's very smart he can do a lot of work but he's limited
1:43:19
he can do a lot of work but he's limited
1:43:19
he can do a lot of work but he's limited by the vehicle upon which he can
1:43:20
by the vehicle upon which he can
1:43:20
by the vehicle upon which he can communicate to the technology and so
1:43:23
communicate to the technology and so
1:43:23
communicate to the technology and so alexa being born a few years ago
1:43:26
alexa being born a few years ago
1:43:26
alexa being born a few years ago really created the opportunity for him
1:43:28
really created the opportunity for him
1:43:28
really created the opportunity for him to do more i also have my dad who's an
1:43:30
to do more i also have my dad who's an
1:43:30
to do more i also have my dad who's an aging
1:43:31
aging
1:43:31
aging you know he's like in his 70s but he
1:43:33
you know he's like in his 70s but he
1:43:33
you know he's like in his 70s but he suffered a traumatic brain injury so
1:43:34
suffered a traumatic brain injury so
1:43:34
suffered a traumatic brain injury so he's got
1:43:35
he's got
1:43:35
he's got a lot of cognitive challenges also not a
1:43:38
a lot of cognitive challenges also not a
1:43:38
a lot of cognitive challenges also not a great candidate for like
1:43:39
great candidate for like
1:43:39
great candidate for like the super smartphone or the keyboard or
1:43:42
the super smartphone or the keyboard or
1:43:42
the super smartphone or the keyboard or the mouse
1:43:43
the mouse
1:43:43
the mouse so i wanted a different world for them
1:43:45
so i wanted a different world for them
1:43:45
so i wanted a different world for them and that's really what drove me
1:43:47
and that's really what drove me
1:43:47
and that's really what drove me to where i am today which is not only
1:43:49
to where i am today which is not only
1:43:49
to where i am today which is not only advocating
1:43:51
advocating
1:43:51
advocating for use cases like this but also for
1:43:54
for use cases like this but also for
1:43:54
for use cases like this but also for technology in general and for
1:43:56
technology in general and for
1:43:56
technology in general and for diversity of thought around all the
1:43:58
diversity of thought around all the
1:43:58
diversity of thought around all the things
1:43:59
things
1:43:59
things so last year and i guess over the last
1:44:01
so last year and i guess over the last
1:44:01
so last year and i guess over the last few years i've gotten lots of accolades
1:44:03
few years i've gotten lots of accolades
1:44:03
few years i've gotten lots of accolades my favorite the reason i love this slide
1:44:05
my favorite the reason i love this slide
1:44:05
my favorite the reason i love this slide is because i got a 3d
1:44:07
is because i got a 3d
1:44:07
is because i got a 3d printed ninja cat my prized award
1:44:10
printed ninja cat my prized award
1:44:10
printed ninja cat my prized award possession
1:44:11
possession
1:44:12
possession um but i got a lot of rewards for just
1:44:14
um but i got a lot of rewards for just
1:44:14
um but i got a lot of rewards for just being willing
1:44:15
being willing
1:44:15
being willing to say out loud we should think
1:44:17
to say out loud we should think
1:44:18
to say out loud we should think differently about the way we're doing
1:44:19
differently about the way we're doing
1:44:19
differently about the way we're doing things a lot
1:44:19
things a lot
1:44:20
things a lot of times as an engineer you just go in
1:44:22
of times as an engineer you just go in
1:44:22
of times as an engineer you just go in and heads down
1:44:23
and heads down
1:44:23
and heads down you solve a problem and you often might
1:44:25
you solve a problem and you often might
1:44:26
you solve a problem and you often might solve it from your own personal
1:44:27
solve it from your own personal
1:44:27
solve it from your own personal perspective but very rarely do i have
1:44:29
perspective but very rarely do i have
1:44:29
perspective but very rarely do i have the luxury as a data scientist
1:44:31
the luxury as a data scientist
1:44:31
the luxury as a data scientist to go talk to the person who originated
1:44:34
to go talk to the person who originated
1:44:34
to go talk to the person who originated the problem or who has the pain
1:44:36
the problem or who has the pain
1:44:36
the problem or who has the pain and that's a lot of what we're going to
1:44:37
and that's a lot of what we're going to
1:44:38
and that's a lot of what we're going to talk about now is how do we really
1:44:39
talk about now is how do we really
1:44:39
talk about now is how do we really create empathy in our engineering teams
1:44:41
create empathy in our engineering teams
1:44:41
create empathy in our engineering teams and what are the tools available to
1:44:43
and what are the tools available to
1:44:43
and what are the tools available to deploy that
1:44:45
deploy that
1:44:45
deploy that um i always like to start with my my
1:44:48
um i always like to start with my my
1:44:48
um i always like to start with my my this quote
1:44:48
this quote
1:44:48
this quote um and the reason why is because it's
1:44:50
um and the reason why is because it's
1:44:50
um and the reason why is because it's actually a data lesson
1:44:52
actually a data lesson
1:44:52
actually a data lesson um so i'll quickly read it to you i know
1:44:54
um so i'll quickly read it to you i know
1:44:54
um so i'll quickly read it to you i know i'm never supposed to do this but those
1:44:55
i'm never supposed to do this but those
1:44:55
i'm never supposed to do this but those of you who've seen me before you know
1:44:57
of you who've seen me before you know
1:44:57
of you who've seen me before you know this is my favorite thing
1:44:58
this is my favorite thing
1:44:58
this is my favorite thing um but what is success to laugh often
1:45:02
um but what is success to laugh often
1:45:02
um but what is success to laugh often and much
1:45:03
and much
1:45:03
and much to win the respective intelligent people
1:45:05
to win the respective intelligent people
1:45:05
to win the respective intelligent people and the affection of children
1:45:07
and the affection of children
1:45:07
and the affection of children to earn the appreciation of honest
1:45:09
to earn the appreciation of honest
1:45:09
to earn the appreciation of honest critics and endure the betrayal of false
1:45:12
critics and endure the betrayal of false
1:45:12
critics and endure the betrayal of false friends
1:45:13
friends
1:45:13
friends to appreciate the beauty to find the
1:45:15
to appreciate the beauty to find the
1:45:15
to appreciate the beauty to find the best in others to leave the world a bit
1:45:17
best in others to leave the world a bit
1:45:17
best in others to leave the world a bit better whether by a healthy child
1:45:19
better whether by a healthy child
1:45:19
better whether by a healthy child or a garden patch or a redeemed social
1:45:22
or a garden patch or a redeemed social
1:45:22
or a garden patch or a redeemed social condition
1:45:23
condition
1:45:23
condition to know even one life has breathed
1:45:25
to know even one life has breathed
1:45:25
to know even one life has breathed easier because you have lived
1:45:28
easier because you have lived
1:45:28
easier because you have lived this is to have succeeded now during my
1:45:30
this is to have succeeded now during my
1:45:30
this is to have succeeded now during my time at amazon
1:45:31
time at amazon
1:45:32
time at amazon i uh jeff bezos actually was the one who
1:45:35
i uh jeff bezos actually was the one who
1:45:35
i uh jeff bezos actually was the one who shared this with me but i've actually
1:45:37
shared this with me but i've actually
1:45:37
shared this with me but i've actually had this quote shared across my
1:45:39
had this quote shared across my
1:45:39
had this quote shared across my technical
1:45:39
technical
1:45:39
technical career and one thing uh that was true
1:45:42
career and one thing uh that was true
1:45:42
career and one thing uh that was true about all of those who
1:45:43
about all of those who
1:45:43
about all of those who who shared it with me was they said it
1:45:45
who shared it with me was they said it
1:45:45
who shared it with me was they said it was written by ralph waldo emerson
1:45:47
was written by ralph waldo emerson
1:45:47
was written by ralph waldo emerson that's who wrote it and after a few
1:45:50
that's who wrote it and after a few
1:45:50
that's who wrote it and after a few sessions of delivering this in global
1:45:51
sessions of delivering this in global
1:45:51
sessions of delivering this in global audiences someone was like
1:45:53
audiences someone was like
1:45:53
audiences someone was like go look that up and i did some research
1:45:56
go look that up and i did some research
1:45:56
go look that up and i did some research and found it actually
1:45:57
and found it actually
1:45:58
and found it actually wasn't amazing it wasn't ralph waldo
1:46:01
wasn't amazing it wasn't ralph waldo
1:46:01
wasn't amazing it wasn't ralph waldo emerson he wrote it who is
1:46:02
emerson he wrote it who is
1:46:02
emerson he wrote it who is you know he definitely made it popular
1:46:04
you know he definitely made it popular
1:46:04
you know he definitely made it popular but it was actually bessie anderson
1:46:06
but it was actually bessie anderson
1:46:06
but it was actually bessie anderson stanley
1:46:07
stanley
1:46:07
stanley and oftentimes there's a lot of things
1:46:09
and oftentimes there's a lot of things
1:46:09
and oftentimes there's a lot of things in our work
1:46:10
in our work
1:46:10
in our work that are presented in a certain way but
1:46:13
that are presented in a certain way but
1:46:13
that are presented in a certain way but upon
1:46:13
upon
1:46:13
upon further kind of digging we find that
1:46:16
further kind of digging we find that
1:46:16
further kind of digging we find that it's not always
1:46:17
it's not always
1:46:17
it's not always the truth and i feel like as a data
1:46:19
the truth and i feel like as a data
1:46:19
the truth and i feel like as a data scientist as a data like an ai
1:46:22
scientist as a data like an ai
1:46:22
scientist as a data like an ai engineer someone who's applying the
1:46:23
engineer someone who's applying the
1:46:23
engineer someone who's applying the models that are becoming democratized
1:46:26
models that are becoming democratized
1:46:26
models that are becoming democratized i want to know i want to dig a little
1:46:28
i want to know i want to dig a little
1:46:28
i want to know i want to dig a little further i want to know
1:46:30
further i want to know
1:46:30
further i want to know what i'm using and why i'm using it and
1:46:33
what i'm using and why i'm using it and
1:46:33
what i'm using and why i'm using it and how i'm using it
1:46:34
how i'm using it
1:46:34
how i'm using it and so a lot of what we're going to talk
1:46:35
and so a lot of what we're going to talk
1:46:35
and so a lot of what we're going to talk about today
1:46:37
about today
1:46:37
about today is about what tools you actually have
1:46:40
is about what tools you actually have
1:46:40
is about what tools you actually have now i always say this you know we have
1:46:42
now i always say this you know we have
1:46:42
now i always say this you know we have more power than ever
1:46:43
more power than ever
1:46:43
more power than ever it's a great time to be alive in tech
1:46:46
it's a great time to be alive in tech
1:46:46
it's a great time to be alive in tech and i
1:46:46
and i
1:46:46
and i always ask this question with great
1:46:49
always ask this question with great
1:46:49
always ask this question with great power right ai gives us immense power
1:46:52
power right ai gives us immense power
1:46:52
power right ai gives us immense power but with great power comes and if you're
1:46:54
but with great power comes and if you're
1:46:54
but with great power comes and if you're in the live chat you can
1:46:55
in the live chat you can
1:46:55
in the live chat you can jot it in there right how do you finish
1:46:57
jot it in there right how do you finish
1:46:57
jot it in there right how do you finish that quote do you know
1:46:59
that quote do you know
1:46:59
that quote do you know uncle ben spiderman with great power
1:47:02
uncle ben spiderman with great power
1:47:02
uncle ben spiderman with great power comes great responsibility and that's
1:47:05
comes great responsibility and that's
1:47:05
comes great responsibility and that's really what this talk is all about
1:47:06
really what this talk is all about
1:47:06
really what this talk is all about is said hopefully you're going to walk
1:47:09
is said hopefully you're going to walk
1:47:09
is said hopefully you're going to walk out of here as a modeler as somebody who
1:47:11
out of here as a modeler as somebody who
1:47:11
out of here as a modeler as somebody who does this work every day
1:47:12
does this work every day
1:47:12
does this work every day and never be able to do that work the
1:47:14
and never be able to do that work the
1:47:14
and never be able to do that work the same way again
1:47:15
same way again
1:47:15
same way again at least that's my hope or at least a
1:47:18
at least that's my hope or at least a
1:47:18
at least that's my hope or at least a spark of thought
1:47:19
spark of thought
1:47:19
spark of thought a new idea will pop into your head when
1:47:21
a new idea will pop into your head when
1:47:22
a new idea will pop into your head when you start doing this work
1:47:23
you start doing this work
1:47:24
you start doing this work so ethics and responsible use of ai is a
1:47:26
so ethics and responsible use of ai is a
1:47:26
so ethics and responsible use of ai is a big
1:47:27
big
1:47:27
big huge topic as a matter of fact fairness
1:47:30
huge topic as a matter of fact fairness
1:47:30
huge topic as a matter of fact fairness in ai is also a gigantic topic i only
1:47:33
in ai is also a gigantic topic i only
1:47:33
in ai is also a gigantic topic i only had a little bit of time
1:47:34
had a little bit of time
1:47:34
had a little bit of time so i'm choosing some very practical
1:47:37
so i'm choosing some very practical
1:47:37
so i'm choosing some very practical things that you can do today to build
1:47:39
things that you can do today to build
1:47:39
things that you can do today to build explainability into your models
1:47:41
explainability into your models
1:47:41
explainability into your models but there are lots of research i mean
1:47:45
but there are lots of research i mean
1:47:45
but there are lots of research i mean i would hundreds of research papers that
1:47:48
i would hundreds of research papers that
1:47:48
i would hundreds of research papers that you could read on different things
1:47:50
you could read on different things
1:47:50
you could read on different things you can do to build a better model and
1:47:52
you can do to build a better model and
1:47:52
you can do to build a better model and that's really the question we're
1:47:53
that's really the question we're
1:47:53
that's really the question we're we're answering right is how do i build
1:47:56
we're answering right is how do i build
1:47:56
we're answering right is how do i build a better model how do i build a model
1:47:58
a better model how do i build a model
1:47:58
a better model how do i build a model that doesn't disenfranchise or doesn't
1:48:01
that doesn't disenfranchise or doesn't
1:48:01
that doesn't disenfranchise or doesn't marginalize and it's hard work and a lot
1:48:04
marginalize and it's hard work and a lot
1:48:04
marginalize and it's hard work and a lot of times our
1:48:05
of times our
1:48:06
of times our teams are not built in a way that would
1:48:08
teams are not built in a way that would
1:48:08
teams are not built in a way that would even recognize it if we were doing that
1:48:10
even recognize it if we were doing that
1:48:10
even recognize it if we were doing that so one of the things i like to always
1:48:12
so one of the things i like to always
1:48:12
so one of the things i like to always mention here with you
1:48:14
mention here with you
1:48:14
mention here with you is look around your team make sure that
1:48:17
is look around your team make sure that
1:48:17
is look around your team make sure that either you
1:48:17
either you
1:48:18
either you have a team that represents a diversity
1:48:20
have a team that represents a diversity
1:48:20
have a team that represents a diversity of thought and i don't just mean gender
1:48:21
of thought and i don't just mean gender
1:48:22
of thought and i don't just mean gender and ethnicity though i do mean that
1:48:24
and ethnicity though i do mean that
1:48:24
and ethnicity though i do mean that but i also mean introverts extroverts
1:48:27
but i also mean introverts extroverts
1:48:27
but i also mean introverts extroverts people with special needs people who are
1:48:28
people with special needs people who are
1:48:28
people with special needs people who are hard of hearing everyone everyone
1:48:32
hard of hearing everyone everyone
1:48:32
hard of hearing everyone everyone is necessary when we are building these
1:48:34
is necessary when we are building these
1:48:34
is necessary when we are building these models because at the end of
1:48:36
models because at the end of
1:48:36
models because at the end of the line in other words when we push a
1:48:38
the line in other words when we push a
1:48:38
the line in other words when we push a model into production
1:48:40
model into production
1:48:40
model into production the reality is is that everyone's going
1:48:42
the reality is is that everyone's going
1:48:42
the reality is is that everyone's going to be using it
1:48:43
to be using it
1:48:43
to be using it and the only way to truly serve those
1:48:45
and the only way to truly serve those
1:48:45
and the only way to truly serve those people we're building for
1:48:46
people we're building for
1:48:46
people we're building for is to make sure they're empathetically
1:48:48
is to make sure they're empathetically
1:48:48
is to make sure they're empathetically or physically
1:48:49
or physically
1:48:49
or physically involved in the development of the
1:48:51
involved in the development of the
1:48:51
involved in the development of the models as we build it
1:48:52
models as we build it
1:48:52
models as we build it so this is all based on this concept of
1:48:55
so this is all based on this concept of
1:48:55
so this is all based on this concept of the principles of ai
1:48:56
the principles of ai
1:48:56
the principles of ai in case you don't know i remember a
1:48:58
in case you don't know i remember a
1:48:58
in case you don't know i remember a couple years ago
1:48:59
couple years ago
1:48:59
couple years ago i think it was at ignite or one of our
1:49:01
i think it was at ignite or one of our
1:49:01
i think it was at ignite or one of our conferences um
1:49:03
conferences um
1:49:03
conferences um the ai teams handed out this really cool
1:49:06
the ai teams handed out this really cool
1:49:06
the ai teams handed out this really cool ruler and it was it was a black ruler
1:49:10
ruler and it was it was a black ruler
1:49:10
ruler and it was it was a black ruler and it had in gold writing the golden
1:49:12
and it had in gold writing the golden
1:49:12
and it had in gold writing the golden rule and it had
1:49:13
rule and it had
1:49:13
rule and it had all of these principles on it and i
1:49:14
all of these principles on it and i
1:49:14
all of these principles on it and i loved it i thought it was such a good
1:49:16
loved it i thought it was such a good
1:49:16
loved it i thought it was such a good pun if you will like what are the rules
1:49:19
pun if you will like what are the rules
1:49:19
pun if you will like what are the rules this is what you should do
1:49:20
this is what you should do
1:49:20
this is what you should do um and i just want to remind you of them
1:49:22
um and i just want to remind you of them
1:49:22
um and i just want to remind you of them we're not going to go through them
1:49:24
we're not going to go through them
1:49:24
we're not going to go through them i just think this is really the under
1:49:26
i just think this is really the under
1:49:26
i just think this is really the under current
1:49:27
current
1:49:27
current of why these tools are becoming more and
1:49:29
of why these tools are becoming more and
1:49:29
of why these tools are becoming more and more prevalent in our organizations and
1:49:31
more prevalent in our organizations and
1:49:31
more prevalent in our organizations and in our conversations so explainable
1:49:35
in our conversations so explainable
1:49:35
in our conversations so explainable i know the title of this was like
1:49:37
i know the title of this was like
1:49:37
i know the title of this was like explainable versus interpretable
1:49:39
explainable versus interpretable
1:49:39
explainable versus interpretable but of course if you know me i am a both
1:49:42
but of course if you know me i am a both
1:49:42
but of course if you know me i am a both and
1:49:43
and
1:49:43
and kind of person so i believe that
1:49:45
kind of person so i believe that
1:49:45
kind of person so i believe that explainability and interpretability
1:49:47
explainability and interpretability
1:49:48
explainability and interpretability are hand in hand that there are really
1:49:50
are hand in hand that there are really
1:49:50
are hand in hand that there are really two ways of describing
1:49:52
two ways of describing
1:49:52
two ways of describing our our interest in making our models
1:49:56
our our interest in making our models
1:49:56
our our interest in making our models understandable in understanding the
1:49:58
understandable in understanding the
1:49:58
understandable in understanding the decisions that they're making
1:50:00
decisions that they're making
1:50:00
decisions that they're making and in my mind to specifically eliminate
1:50:03
and in my mind to specifically eliminate
1:50:03
and in my mind to specifically eliminate bias that shows up when we don't ask
1:50:05
bias that shows up when we don't ask
1:50:05
bias that shows up when we don't ask those questions
1:50:07
those questions
1:50:07
those questions so but we don't even have to worry about
1:50:08
so but we don't even have to worry about
1:50:08
so but we don't even have to worry about bias because if we do the right things
1:50:10
bias because if we do the right things
1:50:10
bias because if we do the right things on the front end
1:50:11
on the front end
1:50:11
on the front end ask the right questions and use the
1:50:13
ask the right questions and use the
1:50:13
ask the right questions and use the right tools which i'm about to show you
1:50:15
right tools which i'm about to show you
1:50:15
right tools which i'm about to show you we don't have to worry about bias
1:50:18
we don't have to worry about bias
1:50:18
we don't have to worry about bias explicitly because by asking those
1:50:21
explicitly because by asking those
1:50:21
explicitly because by asking those questions we will
1:50:22
questions we will
1:50:22
questions we will uncover bias within our models now i do
1:50:24
uncover bias within our models now i do
1:50:24
uncover bias within our models now i do think we have to take care and we have
1:50:26
think we have to take care and we have
1:50:26
think we have to take care and we have to do this
1:50:27
to do this
1:50:27
to do this intentionally but i just mean that these
1:50:29
intentionally but i just mean that these
1:50:29
intentionally but i just mean that these tools will help us do that even if we're
1:50:31
tools will help us do that even if we're
1:50:31
tools will help us do that even if we're in an organization that does not
1:50:33
in an organization that does not
1:50:33
in an organization that does not value those questions being asked so i
1:50:35
value those questions being asked so i
1:50:35
value those questions being asked so i do want to just mention to you there's
1:50:37
do want to just mention to you there's
1:50:37
do want to just mention to you there's this really great tutorial i encourage
1:50:39
this really great tutorial i encourage
1:50:39
this really great tutorial i encourage you all to go
1:50:40
you all to go
1:50:40
you all to go and take a look at it's called i think
1:50:43
and take a look at it's called i think
1:50:43
and take a look at it's called i think it's like
1:50:44
it's like
1:50:44
it's like a explainable ai just google it or bing
1:50:47
a explainable ai just google it or bing
1:50:47
a explainable ai just google it or bing it
1:50:48
it
1:50:48
it um and it'll come up i also put the
1:50:50
um and it'll come up i also put the
1:50:50
um and it'll come up i also put the reference to it at the end of the deck
1:50:52
reference to it at the end of the deck
1:50:52
reference to it at the end of the deck which we'll share with you but there's
1:50:54
which we'll share with you but there's
1:50:54
which we'll share with you but there's three different ways and this
1:50:56
three different ways and this
1:50:56
three different ways and this is so true of my journey in ai i started
1:50:59
is so true of my journey in ai i started
1:51:00
is so true of my journey in ai i started off
1:51:00
off
1:51:00
off really building logic based natural
1:51:03
really building logic based natural
1:51:03
really building logic based natural language
1:51:04
language
1:51:04
language applications um where i was really
1:51:07
applications um where i was really
1:51:07
applications um where i was really creating
1:51:08
creating
1:51:08
creating rules decision trees very simplistic i
1:51:10
rules decision trees very simplistic i
1:51:10
rules decision trees very simplistic i used to call it
1:51:11
used to call it
1:51:11
used to call it weak ai because i was like this isn't
1:51:14
weak ai because i was like this isn't
1:51:14
weak ai because i was like this isn't really a.i
1:51:15
really a.i
1:51:15
really a.i if i'm hard coding every decision
1:51:17
if i'm hard coding every decision
1:51:17
if i'm hard coding every decision that'll ever be made
1:51:19
that'll ever be made
1:51:19
that'll ever be made but there was some it was symbolic ai
1:51:22
but there was some it was symbolic ai
1:51:22
but there was some it was symbolic ai now that i know a little bit more about
1:51:24
now that i know a little bit more about
1:51:24
now that i know a little bit more about that and i
1:51:24
that and i
1:51:24
that and i you know that was seven years ago and
1:51:27
you know that was seven years ago and
1:51:27
you know that was seven years ago and then
1:51:27
then
1:51:28
then some organizations transitioned away
1:51:30
some organizations transitioned away
1:51:30
some organizations transitioned away from that towards more of statistical
1:51:31
from that towards more of statistical
1:51:31
from that towards more of statistical models right and
1:51:33
models right and
1:51:33
models right and identifying these domains for training
1:51:36
identifying these domains for training
1:51:36
identifying these domains for training and using big data to train the model
1:51:38
and using big data to train the model
1:51:38
and using big data to train the model and that's kind of where we are today
1:51:40
and that's kind of where we are today
1:51:40
and that's kind of where we are today right we are for example
1:51:41
right we are for example
1:51:42
right we are for example on a product like alexa or google home
1:51:44
on a product like alexa or google home
1:51:44
on a product like alexa or google home or
1:51:45
or
1:51:45
or any of our naturally natural languages
1:51:47
any of our naturally natural languages
1:51:47
any of our naturally natural languages or even the
1:51:48
or even the
1:51:48
or even the work that anthony just talked about in
1:51:50
work that anthony just talked about in
1:51:50
work that anthony just talked about in training custom vision
1:51:52
training custom vision
1:51:52
training custom vision you're feeding it massive potentially
1:51:54
you're feeding it massive potentially
1:51:54
you're feeding it massive potentially massive amounts of images
1:51:56
massive amounts of images
1:51:56
massive amounts of images in order to train the model to do what
1:51:58
in order to train the model to do what
1:51:58
in order to train the model to do what we want and
1:51:59
we want and
1:51:59
we want and then using statistics probability to
1:52:02
then using statistics probability to
1:52:02
then using statistics probability to determine
1:52:02
determine
1:52:02
determine confidence so we're there what we want
1:52:05
confidence so we're there what we want
1:52:05
confidence so we're there what we want to do now
1:52:06
to do now
1:52:06
to do now is take it to the next level
1:52:09
is take it to the next level
1:52:09
is take it to the next level right gauntlet thrown challenge accepted
1:52:13
right gauntlet thrown challenge accepted
1:52:13
right gauntlet thrown challenge accepted we want to take our statistical models
1:52:15
we want to take our statistical models
1:52:15
we want to take our statistical models and apply
1:52:17
and apply
1:52:17
and apply constructs for explanation and i hope
1:52:20
constructs for explanation and i hope
1:52:20
constructs for explanation and i hope you know when i'm talking to people i
1:52:21
you know when i'm talking to people i
1:52:21
you know when i'm talking to people i used to wear a shirt i'm not wearing it
1:52:23
used to wear a shirt i'm not wearing it
1:52:23
used to wear a shirt i'm not wearing it today that says i heart
1:52:24
today that says i heart
1:52:24
today that says i heart ai and when i would go through the the
1:52:27
ai and when i would go through the the
1:52:27
ai and when i would go through the the airport people would be like
1:52:29
airport people would be like
1:52:29
airport people would be like uh you know that's going to kill us
1:52:31
uh you know that's going to kill us
1:52:31
uh you know that's going to kill us right
1:52:32
right
1:52:32
right you know that's sky skynet right and i'm
1:52:35
you know that's sky skynet right and i'm
1:52:35
you know that's sky skynet right and i'm like
1:52:36
like
1:52:36
like well no not if not if we do do things a
1:52:39
well no not if not if we do do things a
1:52:39
well no not if not if we do do things a little bit differently
1:52:40
little bit differently
1:52:40
little bit differently and i feel like explainable ai is one of
1:52:42
and i feel like explainable ai is one of
1:52:42
and i feel like explainable ai is one of the type one of the methodologies that
1:52:44
the type one of the methodologies that
1:52:44
the type one of the methodologies that you can implement
1:52:45
you can implement
1:52:45
you can implement to protect yourself against these black
1:52:48
to protect yourself against these black
1:52:48
to protect yourself against these black boxes that we might be creating in some
1:52:50
boxes that we might be creating in some
1:52:50
boxes that we might be creating in some of our models
1:52:51
of our models
1:52:52
of our models so really quickly uh i kind of already
1:52:55
so really quickly uh i kind of already
1:52:55
so really quickly uh i kind of already said
1:52:55
said
1:52:55
said this a little bit but just to reiterate
1:52:57
this a little bit but just to reiterate
1:52:57
this a little bit but just to reiterate our current systems
1:52:59
our current systems
1:52:59
our current systems are machine learning centric they are
1:53:01
are machine learning centric they are
1:53:01
are machine learning centric they are statistical and probability based
1:53:04
statistical and probability based
1:53:04
statistical and probability based right they give us confidence numbers
1:53:06
right they give us confidence numbers
1:53:06
right they give us confidence numbers and
1:53:07
and
1:53:07
and oftentimes well i actually always ask
1:53:10
oftentimes well i actually always ask
1:53:10
oftentimes well i actually always ask this
1:53:10
this
1:53:10
this i'll ask this question in just a moment
1:53:12
i'll ask this question in just a moment
1:53:12
i'll ask this question in just a moment but oftentimes in finance in marketing
1:53:14
but oftentimes in finance in marketing
1:53:14
but oftentimes in finance in marketing security in the business
1:53:16
security in the business
1:53:16
security in the business the business will be like great that's
1:53:18
the business will be like great that's
1:53:18
the business will be like great that's awesome how do we do that
1:53:19
awesome how do we do that
1:53:19
awesome how do we do that how do we make that decision um why
1:53:22
how do we make that decision um why
1:53:22
how do we make that decision um why wasn't it this decision
1:53:23
wasn't it this decision
1:53:23
wasn't it this decision why didn't we choose something else and
1:53:25
why didn't we choose something else and
1:53:25
why didn't we choose something else and if we can't easily explain why our model
1:53:28
if we can't easily explain why our model
1:53:28
if we can't easily explain why our model made that decision
1:53:29
made that decision
1:53:29
made that decision it actually creates distrust between the
1:53:31
it actually creates distrust between the
1:53:31
it actually creates distrust between the people we're trying to help
1:53:33
people we're trying to help
1:53:33
people we're trying to help right the modelers i am trying to build
1:53:36
right the modelers i am trying to build
1:53:36
right the modelers i am trying to build a solution that's going to
1:53:37
a solution that's going to
1:53:37
a solution that's going to help a business help a newsroom help
1:53:40
help a business help a newsroom help
1:53:40
help a business help a newsroom help uh you know uh curators at an art
1:53:43
uh you know uh curators at an art
1:53:44
uh you know uh curators at an art gallery right i'm trying to help them
1:53:45
gallery right i'm trying to help them
1:53:45
gallery right i'm trying to help them solve a problem that they have
1:53:47
solve a problem that they have
1:53:47
solve a problem that they have and if i don't do it in a way that's
1:53:48
and if i don't do it in a way that's
1:53:48
and if i don't do it in a way that's explainable it creates distrust because
1:53:51
explainable it creates distrust because
1:53:51
explainable it creates distrust because they don't understand
1:53:52
they don't understand
1:53:52
they don't understand why or how that decision is being made
1:53:55
why or how that decision is being made
1:53:55
why or how that decision is being made so it makes it really important not only
1:53:57
so it makes it really important not only
1:53:57
so it makes it really important not only for us
1:53:57
for us
1:53:57
for us because we want to be able to eliminate
1:53:59
because we want to be able to eliminate
1:53:59
because we want to be able to eliminate bias which is a nice
1:54:01
bias which is a nice
1:54:01
bias which is a nice reason um an incredibly important reason
1:54:03
reason um an incredibly important reason
1:54:04
reason um an incredibly important reason but also because
1:54:05
but also because
1:54:05
but also because we're serving the community with this
1:54:06
we're serving the community with this
1:54:06
we're serving the community with this technology we're building models that
1:54:08
technology we're building models that
1:54:08
technology we're building models that serve people and if we want to serve
1:54:10
serve people and if we want to serve
1:54:10
serve people and if we want to serve people getting them to believe
1:54:13
people getting them to believe
1:54:13
people getting them to believe invest and understand what we're doing
1:54:15
invest and understand what we're doing
1:54:15
invest and understand what we're doing is part of that puzzle
1:54:17
is part of that puzzle
1:54:17
is part of that puzzle and so explainability helps in that way
1:54:19
and so explainability helps in that way
1:54:19
and so explainability helps in that way as well
1:54:20
as well
1:54:20
as well so why do we even need this just to
1:54:23
so why do we even need this just to
1:54:23
so why do we even need this just to reiterate right
1:54:24
reiterate right
1:54:24
reiterate right really it comes down to one thing bad
1:54:26
really it comes down to one thing bad
1:54:26
really it comes down to one thing bad decisions and ai models are dangerous
1:54:29
decisions and ai models are dangerous
1:54:29
decisions and ai models are dangerous so i always like to ask that you know
1:54:31
so i always like to ask that you know
1:54:31
so i always like to ask that you know you all that are listening can you think
1:54:32
you all that are listening can you think
1:54:32
you all that are listening can you think of an example of an ai model gone bad
1:54:37
of an example of an ai model gone bad
1:54:37
of an example of an ai model gone bad i you know gone wild gone rogue that has
1:54:39
i you know gone wild gone rogue that has
1:54:40
i you know gone wild gone rogue that has made
1:54:40
made
1:54:40
made a bad decision go ahead and jot it in
1:54:42
a bad decision go ahead and jot it in
1:54:42
a bad decision go ahead and jot it in the chat there's lots of examples
1:54:44
the chat there's lots of examples
1:54:44
the chat there's lots of examples because we didn't ask these questions
1:54:46
because we didn't ask these questions
1:54:46
because we didn't ask these questions even five years ago models that went
1:54:48
even five years ago models that went
1:54:48
even five years ago models that went into
1:54:49
into
1:54:49
into you know models that were deployed or
1:54:50
you know models that were deployed or
1:54:50
you know models that were deployed or went into production especially in the
1:54:53
went into production especially in the
1:54:53
went into production especially in the startup world
1:54:54
startup world
1:54:54
startup world we had challenges we're now even seeing
1:54:56
we had challenges we're now even seeing
1:54:56
we had challenges we're now even seeing companies like
1:54:58
companies like
1:54:58
companies like amazon and ibm pull out of
1:55:01
amazon and ibm pull out of
1:55:01
amazon and ibm pull out of facial recognition because they're like
1:55:03
facial recognition because they're like
1:55:03
facial recognition because they're like this is not good
1:55:04
this is not good
1:55:04
this is not good these models are making bad choices um
1:55:07
these models are making bad choices um
1:55:07
these models are making bad choices um so
1:55:08
so
1:55:08
so bad decisions can be dangerous
1:55:10
bad decisions can be dangerous
1:55:10
bad decisions can be dangerous especially in the areas
1:55:11
especially in the areas
1:55:11
especially in the areas of right i'm hopefully someone thought
1:55:14
of right i'm hopefully someone thought
1:55:14
of right i'm hopefully someone thought about autonomous vehicles
1:55:16
about autonomous vehicles
1:55:16
about autonomous vehicles that's not good right or finance
1:55:19
that's not good right or finance
1:55:19
that's not good right or finance credit decisions or i don't know if
1:55:21
credit decisions or i don't know if
1:55:21
credit decisions or i don't know if you've heard about the one where
1:55:22
you've heard about the one where
1:55:22
you've heard about the one where somebody created one around
1:55:24
somebody created one around
1:55:24
somebody created one around judgments like in the actual judicial
1:55:26
judgments like in the actual judicial
1:55:26
judgments like in the actual judicial system
1:55:27
system
1:55:27
system or in the prison system when you have
1:55:29
or in the prison system when you have
1:55:29
or in the prison system when you have bad decisions in your model it doesn't
1:55:31
bad decisions in your model it doesn't
1:55:31
bad decisions in your model it doesn't mean your model's bad
1:55:33
mean your model's bad
1:55:33
mean your model's bad it just means it doesn't have enough
1:55:35
it just means it doesn't have enough
1:55:35
it just means it doesn't have enough data it doesn't have a diverse enough
1:55:37
data it doesn't have a diverse enough
1:55:37
data it doesn't have a diverse enough set of data
1:55:38
set of data
1:55:38
set of data it doesn't have enough information to
1:55:39
it doesn't have enough information to
1:55:39
it doesn't have enough information to make the right decision
1:55:41
make the right decision
1:55:41
make the right decision and so having explainability though
1:55:44
and so having explainability though
1:55:44
and so having explainability though allows us to tease out those bad
1:55:46
allows us to tease out those bad
1:55:46
allows us to tease out those bad decisions
1:55:46
decisions
1:55:46
decisions figure out why the model made that
1:55:49
figure out why the model made that
1:55:49
figure out why the model made that decision
1:55:50
decision
1:55:50
decision and train it to make a better one and
1:55:52
and train it to make a better one and
1:55:52
and train it to make a better one and that really
1:55:53
that really
1:55:53
that really is that's why i always say never give up
1:55:55
is that's why i always say never give up
1:55:55
is that's why i always say never give up on a model
1:55:56
on a model
1:55:56
on a model okay that's not true in ai i try to
1:55:59
okay that's not true in ai i try to
1:55:59
okay that's not true in ai i try to avoid
1:56:00
avoid
1:56:00
avoid absolutes like always and never um
1:56:03
absolutes like always and never um
1:56:03
absolutes like always and never um but be very hesitant to give up on a
1:56:05
but be very hesitant to give up on a
1:56:06
but be very hesitant to give up on a model like
1:56:06
model like
1:56:06
model like some we're starting to see some
1:56:08
some we're starting to see some
1:56:08
some we're starting to see some companies decide to do
1:56:10
companies decide to do
1:56:10
companies decide to do when the ch when the solution might
1:56:12
when the ch when the solution might
1:56:12
when the ch when the solution might actually be
1:56:13
actually be
1:56:13
actually be just spend a little time be intentional
1:56:15
just spend a little time be intentional
1:56:16
just spend a little time be intentional about your approach
1:56:17
about your approach
1:56:17
about your approach and take the time to debug
1:56:20
and take the time to debug
1:56:20
and take the time to debug your models so that's really kind of
1:56:22
your models so that's really kind of
1:56:22
your models so that's really kind of what we're talking about here
1:56:23
what we're talking about here
1:56:23
what we're talking about here are different mechanisms you can use
1:56:25
are different mechanisms you can use
1:56:25
are different mechanisms you can use them use as a modeler
1:56:27
them use as a modeler
1:56:27
them use as a modeler to debug your solutions so here's an
1:56:30
to debug your solutions so here's an
1:56:30
to debug your solutions so here's an image
1:56:31
image
1:56:31
image um and this is really what scopes our
1:56:33
um and this is really what scopes our
1:56:33
um and this is really what scopes our conversation today because i think
1:56:35
conversation today because i think
1:56:35
conversation today because i think a lot and i'm asked a lot by executives
1:56:38
a lot and i'm asked a lot by executives
1:56:38
a lot and i'm asked a lot by executives about
1:56:39
about
1:56:39
about like how are we going to do this and
1:56:41
like how are we going to do this and
1:56:41
like how are we going to do this and oftentimes
1:56:42
oftentimes
1:56:42
oftentimes um and a lot of people are like do you
1:56:44
um and a lot of people are like do you
1:56:44
um and a lot of people are like do you let an ai model make
1:56:46
let an ai model make
1:56:46
let an ai model make 100 of the decision now those of us in
1:56:48
100 of the decision now those of us in
1:56:48
100 of the decision now those of us in this industry know
1:56:49
this industry know
1:56:49
this industry know that that's usually not the case we have
1:56:52
that that's usually not the case we have
1:56:52
that that's usually not the case we have today
1:56:52
today
1:56:52
today on the left hand side right this or
1:56:54
on the left hand side right this or
1:56:54
on the left hand side right this or maybe it's the right hand for you i
1:56:55
maybe it's the right hand for you i
1:56:55
maybe it's the right hand for you i don't know
1:56:57
don't know
1:56:57
don't know standard ml right where we've got data
1:56:59
standard ml right where we've got data
1:56:59
standard ml right where we've got data we ingest data into a model
1:57:01
we ingest data into a model
1:57:01
we ingest data into a model we allow that model to make predictions
1:57:03
we allow that model to make predictions
1:57:03
we allow that model to make predictions and then there's some errors that occur
1:57:05
and then there's some errors that occur
1:57:05
and then there's some errors that occur what we want to do is have that same
1:57:08
what we want to do is have that same
1:57:08
what we want to do is have that same process happen but instead inject
1:57:10
process happen but instead inject
1:57:10
process happen but instead inject interpretability the ability to see into
1:57:13
interpretability the ability to see into
1:57:14
interpretability the ability to see into how those decisions are being made and
1:57:16
how those decisions are being made and
1:57:16
how those decisions are being made and then let a human
1:57:17
then let a human
1:57:18
then let a human like me look at that data and this is
1:57:20
like me look at that data and this is
1:57:20
like me look at that data and this is similar to what is done in custom vision
1:57:22
similar to what is done in custom vision
1:57:22
similar to what is done in custom vision for example
1:57:23
for example
1:57:23
for example i go in and i look and i go why do you
1:57:26
i go in and i look and i go why do you
1:57:26
i go in and i look and i go why do you think that that's a wrench
1:57:28
think that that's a wrench
1:57:28
think that that's a wrench when it's actually you know needle nose
1:57:30
when it's actually you know needle nose
1:57:30
when it's actually you know needle nose pliers
1:57:31
pliers
1:57:31
pliers why do you think that let me go in and
1:57:33
why do you think that let me go in and
1:57:33
why do you think that let me go in and figure it out it requires
1:57:35
figure it out it requires
1:57:35
figure it out it requires human inspection but the nice thing
1:57:37
human inspection but the nice thing
1:57:37
human inspection but the nice thing about that
1:57:38
about that
1:57:38
about that is that when you use human inspection
1:57:39
is that when you use human inspection
1:57:40
is that when you use human inspection you gain additional insight that can be
1:57:42
you gain additional insight that can be
1:57:42
you gain additional insight that can be documented along with your model
1:57:44
documented along with your model
1:57:44
documented along with your model increasing its explainability so as it
1:57:46
increasing its explainability so as it
1:57:46
increasing its explainability so as it says here at the top i believe that this
1:57:48
says here at the top i believe that this
1:57:48
says here at the top i believe that this is a both and
1:57:50
is a both and
1:57:50
is a both and solution so let's talk through a couple
1:57:52
solution so let's talk through a couple
1:57:52
solution so let's talk through a couple different examples
1:57:53
different examples
1:57:53
different examples i call these human approaches to
1:57:55
i call these human approaches to
1:57:55
i call these human approaches to explainability because there
1:57:57
explainability because there
1:57:57
explainability because there are packages that are available i'll
1:57:59
are packages that are available i'll
1:57:59
are packages that are available i'll mention them at the end
1:58:00
mention them at the end
1:58:00
mention them at the end that you can use to actually basically
1:58:03
that you can use to actually basically
1:58:03
that you can use to actually basically go through your models and extract out
1:58:05
go through your models and extract out
1:58:05
go through your models and extract out things that are potential concerns
1:58:07
things that are potential concerns
1:58:07
things that are potential concerns i like the crawl walk run philosophy
1:58:10
i like the crawl walk run philosophy
1:58:10
i like the crawl walk run philosophy it's how i learn to code
1:58:12
it's how i learn to code
1:58:12
it's how i learn to code where i base i do everything by myself
1:58:15
where i base i do everything by myself
1:58:15
where i base i do everything by myself first
1:58:15
first
1:58:15
first i write the model from scratch first
1:58:18
i write the model from scratch first
1:58:18
i write the model from scratch first then i use an applied ai model
1:58:19
then i use an applied ai model
1:58:20
then i use an applied ai model in this case i like the idea that we
1:58:22
in this case i like the idea that we
1:58:22
in this case i like the idea that we learn how our models
1:58:23
learn how our models
1:58:23
learn how our models work manually like we go in and we look
1:58:26
work manually like we go in and we look
1:58:26
work manually like we go in and we look we do
1:58:27
we do
1:58:27
we do what i'm about to show you so do walk
1:58:29
what i'm about to show you so do walk
1:58:29
what i'm about to show you so do walk through these three approaches
1:58:31
through these three approaches
1:58:31
through these three approaches and then we go ahead and automate those
1:58:34
and then we go ahead and automate those
1:58:34
and then we go ahead and automate those processes with packages and tool sets
1:58:37
processes with packages and tool sets
1:58:37
processes with packages and tool sets but i do think there's value in really
1:58:39
but i do think there's value in really
1:58:39
but i do think there's value in really understanding how this works
1:58:40
understanding how this works
1:58:40
understanding how this works um yourself before using a tool to
1:58:43
um yourself before using a tool to
1:58:43
um yourself before using a tool to automate it
1:58:44
automate it
1:58:44
automate it so the first approach and i'll go
1:58:46
so the first approach and i'll go
1:58:46
so the first approach and i'll go through these rather quickly we are
1:58:47
through these rather quickly we are
1:58:47
through these rather quickly we are going to have a panel
1:58:49
going to have a panel
1:58:49
going to have a panel and i believe right after me you're
1:58:51
and i believe right after me you're
1:58:51
and i believe right after me you're going to have a really nice deep dive on
1:58:54
going to have a really nice deep dive on
1:58:54
going to have a really nice deep dive on pie torch and how to build
1:58:55
pie torch and how to build
1:58:55
pie torch and how to build explainability there so it's going to be
1:58:57
explainability there so it's going to be
1:58:57
explainability there so it's going to be really exciting talk so i hope you'll
1:58:58
really exciting talk so i hope you'll
1:58:58
really exciting talk so i hope you'll join us
1:58:59
join us
1:58:59
join us um but in the first i have three
1:59:01
um but in the first i have three
1:59:01
um but in the first i have three approaches i want to quickly go over
1:59:03
approaches i want to quickly go over
1:59:03
approaches i want to quickly go over that i feel like i can cover in my short
1:59:05
that i feel like i can cover in my short
1:59:05
that i feel like i can cover in my short amount of time
1:59:06
amount of time
1:59:06
amount of time in enough detail that you can start
1:59:08
in enough detail that you can start
1:59:08
in enough detail that you can start using it immediately
1:59:09
using it immediately
1:59:09
using it immediately one is post talk after
1:59:12
one is post talk after
1:59:12
one is post talk after development being able to just explain
1:59:14
development being able to just explain
1:59:14
development being able to just explain your model so this
1:59:15
your model so this
1:59:16
your model so this is literally like whiteboarding it out
1:59:18
is literally like whiteboarding it out
1:59:18
is literally like whiteboarding it out being able to present it
1:59:20
being able to present it
1:59:20
being able to present it to a business stakeholder oftentimes
1:59:23
to a business stakeholder oftentimes
1:59:23
to a business stakeholder oftentimes when i present
1:59:24
when i present
1:59:24
when i present a stage of development for an ai model
1:59:27
a stage of development for an ai model
1:59:27
a stage of development for an ai model i'm asked enough questions in that that
1:59:29
i'm asked enough questions in that that
1:59:29
i'm asked enough questions in that that it forces me
1:59:30
it forces me
1:59:30
it forces me to go back and create explainability in
1:59:33
to go back and create explainability in
1:59:33
to go back and create explainability in my model
1:59:34
my model
1:59:34
my model this one in my mind is kind of i don't
1:59:37
this one in my mind is kind of i don't
1:59:37
this one in my mind is kind of i don't know
1:59:37
know
1:59:37
know somewhat the easier way to go because it
1:59:40
somewhat the easier way to go because it
1:59:40
somewhat the easier way to go because it is you
1:59:41
is you
1:59:41
is you looking and understanding a model that
1:59:43
looking and understanding a model that
1:59:43
looking and understanding a model that hopefully you created
1:59:44
hopefully you created
1:59:44
hopefully you created and then defining and creating
1:59:47
and then defining and creating
1:59:47
and then defining and creating explanations for that so
1:59:48
explanations for that so
1:59:48
explanations for that so hopefully this is self-explanatory let's
1:59:50
hopefully this is self-explanatory let's
1:59:50
hopefully this is self-explanatory let's get into some more complicated versions
1:59:53
get into some more complicated versions
1:59:53
get into some more complicated versions ablations are another one's actually
1:59:55
ablations are another one's actually
1:59:55
ablations are another one's actually taken up from the surgical world well
1:59:57
taken up from the surgical world well
1:59:57
taken up from the surgical world well maybe it's not even original to them but
1:59:59
maybe it's not even original to them but
1:59:59
maybe it's not even original to them but this is where you drop a future
2:00:02
this is where you drop a future
2:00:02
this is where you drop a future right and an attribute and you change
2:00:04
right and an attribute and you change
2:00:04
right and an attribute and you change the prediction of it
2:00:06
the prediction of it
2:00:06
the prediction of it in order to actually see how the model
2:00:08
in order to actually see how the model
2:00:08
in order to actually see how the model reacts
2:00:09
reacts
2:00:09
reacts now before we get into this i actually
2:00:12
now before we get into this i actually
2:00:12
now before we get into this i actually want to see if i can
2:00:14
want to see if i can
2:00:14
want to see if i can let me i'm going to escape out of my
2:00:16
let me i'm going to escape out of my
2:00:16
let me i'm going to escape out of my presentation
2:00:17
presentation
2:00:17
presentation and just show you a way of testing out
2:00:20
and just show you a way of testing out
2:00:20
and just show you a way of testing out this concept of feature identification
2:00:24
this concept of feature identification
2:00:24
this concept of feature identification and i'm just going gonna do it right in
2:00:25
and i'm just going gonna do it right in
2:00:25
and i'm just going gonna do it right in the browser because i know we have
2:00:28
the browser because i know we have
2:00:28
the browser because i know we have oh dear did i am i in the right place
2:00:30
oh dear did i am i in the right place
2:00:30
oh dear did i am i in the right place let me go back
2:00:31
let me go back
2:00:31
let me go back um so i just want to test it out right
2:00:33
um so i just want to test it out right
2:00:33
um so i just want to test it out right in the browser so you can see what it
2:00:35
in the browser so you can see what it
2:00:35
in the browser so you can see what it looks like
2:00:35
looks like
2:00:35
looks like it's really kind of neat that you could
2:00:37
it's really kind of neat that you could
2:00:37
it's really kind of neat that you could do this in my mind um
2:00:39
do this in my mind um
2:00:39
do this in my mind um and it's to make a point i mean i don't
2:00:41
and it's to make a point i mean i don't
2:00:41
and it's to make a point i mean i don't expect you all to go here i would hope
2:00:43
expect you all to go here i would hope
2:00:43
expect you all to go here i would hope that you just build it out in your
2:00:44
that you just build it out in your
2:00:44
that you just build it out in your development environment
2:00:46
development environment
2:00:46
development environment but if i go into computer vision right
2:00:48
but if i go into computer vision right
2:00:48
but if i go into computer vision right here it allows me to
2:00:50
here it allows me to
2:00:50
here it allows me to submit a image into
2:00:53
submit a image into
2:00:53
submit a image into you'll see down here right into a model
2:00:55
you'll see down here right into a model
2:00:55
you'll see down here right into a model and get metadata back
2:00:57
and get metadata back
2:00:57
and get metadata back so the nice thing about doing this and
2:00:59
so the nice thing about doing this and
2:00:59
so the nice thing about doing this and i'm just gonna upload an image
2:01:01
i'm just gonna upload an image
2:01:01
i'm just gonna upload an image um i i don't know if you all saw it but
2:01:03
um i i don't know if you all saw it but
2:01:03
um i i don't know if you all saw it but about two years ago i did a computer
2:01:05
about two years ago i did a computer
2:01:05
about two years ago i did a computer vision
2:01:06
vision
2:01:06
vision a custom vision example on a very famous
2:01:09
a custom vision example on a very famous
2:01:09
a custom vision example on a very famous character set uh known as i have to be
2:01:12
character set uh known as i have to be
2:01:12
character set uh known as i have to be very careful here
2:01:13
very careful here
2:01:13
very careful here chip and dale chipmunks from disney
2:01:17
chip and dale chipmunks from disney
2:01:17
chip and dale chipmunks from disney and when i choose that you'll notice
2:01:21
and when i choose that you'll notice
2:01:21
and when i choose that you'll notice the metadata that the model gives me
2:01:22
the metadata that the model gives me
2:01:22
the metadata that the model gives me back is like
2:01:24
back is like
2:01:24
back is like teddy bear cute nice a toy
2:01:28
teddy bear cute nice a toy
2:01:28
teddy bear cute nice a toy um so this isn't right in case you don't
2:01:31
um so this isn't right in case you don't
2:01:31
um so this isn't right in case you don't know
2:01:32
know
2:01:32
know and so what i want to do is what we want
2:01:35
and so what i want to do is what we want
2:01:35
and so what i want to do is what we want to do
2:01:36
to do
2:01:36
to do as we are building explainability into
2:01:37
as we are building explainability into
2:01:37
as we are building explainability into our models is go through these examples
2:01:40
our models is go through these examples
2:01:40
our models is go through these examples identify where the errors are within
2:01:42
identify where the errors are within
2:01:42
identify where the errors are within these decisions
2:01:43
these decisions
2:01:43
these decisions and find out why that error exists right
2:01:46
and find out why that error exists right
2:01:46
and find out why that error exists right like it's 74
2:01:48
like it's 74
2:01:48
like it's 74 confident that this is a toy that's
2:01:51
confident that this is a toy that's
2:01:51
confident that this is a toy that's relatively high
2:01:52
relatively high
2:01:52
relatively high so i want to talk i want to be able to
2:01:54
so i want to talk i want to be able to
2:01:54
so i want to talk i want to be able to explain why that happened and then most
2:01:56
explain why that happened and then most
2:01:56
explain why that happened and then most importantly
2:01:57
importantly
2:01:57
importantly how do i train it on a way like how do i
2:01:59
how do i train it on a way like how do i
2:01:59
how do i train it on a way like how do i train this model so it makes a different
2:02:01
train this model so it makes a different
2:02:01
train this model so it makes a different decision
2:02:02
decision
2:02:02
decision with custom vision it's kind of nice
2:02:03
with custom vision it's kind of nice
2:02:03
with custom vision it's kind of nice because i could do all of that without
2:02:05
because i could do all of that without
2:02:05
because i could do all of that without code
2:02:05
code
2:02:05
code but it doesn't really matter it all
2:02:06
but it doesn't really matter it all
2:02:06
but it doesn't really matter it all comes down to labeling i want to be able
2:02:08
comes down to labeling i want to be able
2:02:08
comes down to labeling i want to be able to go in and create new labels that i
2:02:10
to go in and create new labels that i
2:02:10
to go in and create new labels that i identify what i actually want the model
2:02:12
identify what i actually want the model
2:02:12
identify what i actually want the model to understand
2:02:13
to understand
2:02:13
to understand as opposed to it defaulting to some
2:02:16
as opposed to it defaulting to some
2:02:16
as opposed to it defaulting to some level of truth that it got from a
2:02:17
level of truth that it got from a
2:02:17
level of truth that it got from a thousand
2:02:18
thousand
2:02:18
thousand other images that looked kind of similar
2:02:21
other images that looked kind of similar
2:02:21
other images that looked kind of similar now in the second slide that i just
2:02:23
now in the second slide that i just
2:02:23
now in the second slide that i just showed you i also talked about
2:02:25
showed you i also talked about
2:02:25
showed you i also talked about gradients and gradients against a
2:02:28
gradients and gradients against a
2:02:28
gradients and gradients against a baseline
2:02:29
baseline
2:02:29
baseline so when you're building explainability
2:02:30
so when you're building explainability
2:02:30
so when you're building explainability into your model baselines are really
2:02:33
into your model baselines are really
2:02:33
into your model baselines are really important especially as you talk about
2:02:35
important especially as you talk about
2:02:35
important especially as you talk about attributions
2:02:36
attributions
2:02:36
attributions and so you want to establish a baseline
2:02:38
and so you want to establish a baseline
2:02:38
and so you want to establish a baseline of like this is chip and dale
2:02:40
of like this is chip and dale
2:02:40
of like this is chip and dale these are chipmunks they are disney
2:02:42
these are chipmunks they are disney
2:02:42
these are chipmunks they are disney characters they are in costume
2:02:45
characters they are in costume
2:02:45
characters they are in costume you want to establish a baseline of
2:02:46
you want to establish a baseline of
2:02:46
you want to establish a baseline of images that is the ground truth of your
2:02:49
images that is the ground truth of your
2:02:49
images that is the ground truth of your model
2:02:50
model
2:02:50
model and that way as you go through these
2:02:52
and that way as you go through these
2:02:52
and that way as you go through these tests and you get
2:02:53
tests and you get
2:02:53
tests and you get failure results it's easy to map it to
2:02:56
failure results it's easy to map it to
2:02:56
failure results it's easy to map it to some sort of ground truth
2:02:58
some sort of ground truth
2:02:58
some sort of ground truth but in addition to that you can train
2:03:00
but in addition to that you can train
2:03:00
but in addition to that you can train your model on
2:03:01
your model on
2:03:01
your model on gradients right on feature gradients so
2:03:04
gradients right on feature gradients so
2:03:04
gradients right on feature gradients so you take a feature
2:03:05
you take a feature
2:03:05
you take a feature and you provide especially in custom
2:03:06
and you provide especially in custom
2:03:06
and you provide especially in custom vision it's kind of easy to think about
2:03:08
vision it's kind of easy to think about
2:03:08
vision it's kind of easy to think about if you're thinking about the gradients
2:03:10
if you're thinking about the gradients
2:03:10
if you're thinking about the gradients of this image
2:03:11
of this image
2:03:11
of this image i might have one that is super dark one
2:03:14
i might have one that is super dark one
2:03:14
i might have one that is super dark one that's super light
2:03:15
that's super light
2:03:15
that's super light one that um has high contrast low
2:03:18
one that um has high contrast low
2:03:18
one that um has high contrast low contrast right different gradients
2:03:20
contrast right different gradients
2:03:20
contrast right different gradients of um clarity on the image and train it
2:03:23
of um clarity on the image and train it
2:03:23
of um clarity on the image and train it on the exact same labels
2:03:25
on the exact same labels
2:03:25
on the exact same labels so that i can get the nuanced decision
2:03:28
so that i can get the nuanced decision
2:03:28
so that i can get the nuanced decision making
2:03:29
making
2:03:29
making um to be more accurate as the model
2:03:31
um to be more accurate as the model
2:03:31
um to be more accurate as the model starts to evaluate these images
2:03:33
starts to evaluate these images
2:03:33
starts to evaluate these images so it's just kind of i like uh being
2:03:35
so it's just kind of i like uh being
2:03:35
so it's just kind of i like uh being able to just go in here
2:03:37
able to just go in here
2:03:37
able to just go in here throw up an image and see what this
2:03:38
throw up an image and see what this
2:03:38
throw up an image and see what this resnet chain model would give me
2:03:41
resnet chain model would give me
2:03:41
resnet chain model would give me but you could do the same thing with
2:03:42
but you could do the same thing with
2:03:42
but you could do the same thing with your own model this is easier to do in a
2:03:44
your own model this is easier to do in a
2:03:44
your own model this is easier to do in a lecture type environment
2:03:46
lecture type environment
2:03:46
lecture type environment um but something that you can do for
2:03:47
um but something that you can do for
2:03:47
um but something that you can do for yourself as well so i think
2:03:49
yourself as well so i think
2:03:50
yourself as well so i think i just uh oh let me i'll just go back
2:03:53
i just uh oh let me i'll just go back
2:03:53
i just uh oh let me i'll just go back into presentation mode really quick
2:03:55
into presentation mode really quick
2:03:55
into presentation mode really quick so being able to go in and then look at
2:03:57
so being able to go in and then look at
2:03:57
so being able to go in and then look at the features that i'm
2:03:58
the features that i'm
2:03:58
the features that i'm identifying and drop one and see how my
2:04:00
identifying and drop one and see how my
2:04:00
identifying and drop one and see how my predictions change will also help
2:04:02
predictions change will also help
2:04:02
predictions change will also help uncover this
2:04:03
uncover this
2:04:03
uncover this the only challenge with this is that at
2:04:05
the only challenge with this is that at
2:04:05
the only challenge with this is that at least
2:04:06
least
2:04:06
least that i've experienced is the level of
2:04:08
that i've experienced is the level of
2:04:08
that i've experienced is the level of compute is
2:04:09
compute is
2:04:09
compute is it's pretty expensive in other words you
2:04:12
it's pretty expensive in other words you
2:04:12
it's pretty expensive in other words you have to process lots and lots and lots
2:04:13
have to process lots and lots and lots
2:04:14
have to process lots and lots and lots of these
2:04:14
of these
2:04:14
of these images so i also like this idea
2:04:18
images so i also like this idea
2:04:18
images so i also like this idea and i would be remiss if i didn't
2:04:19
and i would be remiss if i didn't
2:04:19
and i would be remiss if i didn't mention the studies
2:04:21
mention the studies
2:04:21
mention the studies that support these human intervention
2:04:24
that support these human intervention
2:04:24
that support these human intervention ideas
2:04:24
ideas
2:04:24
ideas around um modeling uh feature accents
2:04:28
around um modeling uh feature accents
2:04:28
around um modeling uh feature accents or modeling uh or building
2:04:31
or modeling uh or building
2:04:31
or modeling uh or building explainability into your model
2:04:33
explainability into your model
2:04:33
explainability into your model but in this specific example we're
2:04:35
but in this specific example we're
2:04:35
but in this specific example we're talking about like looking the
2:04:36
talking about like looking the
2:04:36
talking about like looking the prediction that's going to be made
2:04:38
prediction that's going to be made
2:04:38
prediction that's going to be made and now instead of just like what i
2:04:41
and now instead of just like what i
2:04:41
and now instead of just like what i would have done
2:04:42
would have done
2:04:42
would have done in a computer vision or custom vision
2:04:44
in a computer vision or custom vision
2:04:44
in a computer vision or custom vision scenario
2:04:45
scenario
2:04:45
scenario and just extracted a feature and seen
2:04:47
and just extracted a feature and seen
2:04:47
and just extracted a feature and seen what happens to the prediction
2:04:48
what happens to the prediction
2:04:48
what happens to the prediction i can actually take a sample of data and
2:04:51
i can actually take a sample of data and
2:04:51
i can actually take a sample of data and do that a very similar exercise but
2:04:53
do that a very similar exercise but
2:04:53
do that a very similar exercise but against a sample
2:04:54
against a sample
2:04:54
against a sample and that will give me a slightly
2:04:55
and that will give me a slightly
2:04:55
and that will give me a slightly different result but it will allow me
2:04:57
different result but it will allow me
2:04:57
different result but it will allow me to as a human as the uh modeler
2:05:01
to as a human as the uh modeler
2:05:01
to as a human as the uh modeler to discern why those decisions are the
2:05:04
to discern why those decisions are the
2:05:04
to discern why those decisions are the way they
2:05:04
way they
2:05:04
way they are um and so in this you know scope of
2:05:08
are um and so in this you know scope of
2:05:08
are um and so in this you know scope of a few minutes that i have i i think
2:05:11
a few minutes that i have i i think
2:05:11
a few minutes that i have i i think these three things
2:05:12
these three things
2:05:12
these three things if you can do nothing just start by
2:05:14
if you can do nothing just start by
2:05:14
if you can do nothing just start by being able to explain the decisions your
2:05:16
being able to explain the decisions your
2:05:16
being able to explain the decisions your models can
2:05:17
models can
2:05:17
models can are making right just start there set
2:05:20
are making right just start there set
2:05:20
are making right just start there set the intention
2:05:21
the intention
2:05:21
the intention when you build a model that you're going
2:05:22
when you build a model that you're going
2:05:22
when you build a model that you're going to ask the question
2:05:24
to ask the question
2:05:24
to ask the question why does this happen and then have an
2:05:27
why does this happen and then have an
2:05:27
why does this happen and then have an answer
2:05:27
answer
2:05:28
answer you could have that answer because you
2:05:29
you could have that answer because you
2:05:29
you could have that answer because you just know but in
2:05:31
just know but in
2:05:31
just know but in deep learning or in neural networks it
2:05:34
deep learning or in neural networks it
2:05:34
deep learning or in neural networks it gets
2:05:34
gets
2:05:34
gets exponentially more complex to create
2:05:36
exponentially more complex to create
2:05:36
exponentially more complex to create that explainability
2:05:37
that explainability
2:05:38
that explainability off the top of your head so that's where
2:05:40
off the top of your head so that's where
2:05:40
off the top of your head so that's where tools come in
2:05:41
tools come in
2:05:41
tools come in tools and resources come in and we want
2:05:43
tools and resources come in and we want
2:05:43
tools and resources come in and we want to really
2:05:44
to really
2:05:44
to really again do a combination of both make sure
2:05:46
again do a combination of both make sure
2:05:46
again do a combination of both make sure you're asking the right questions
2:05:48
you're asking the right questions
2:05:48
you're asking the right questions driving towards those answers as you're
2:05:51
driving towards those answers as you're
2:05:51
driving towards those answers as you're building your models
2:05:52
building your models
2:05:52
building your models but then also leverage some tools
2:05:55
but then also leverage some tools
2:05:55
but then also leverage some tools so i put up here some resources that the
2:05:58
so i put up here some resources that the
2:05:58
so i put up here some resources that the team at global ai shared with me that i
2:06:00
team at global ai shared with me that i
2:06:00
team at global ai shared with me that i thought were amazing and i had not read
2:06:02
thought were amazing and i had not read
2:06:02
thought were amazing and i had not read prior to joining but also some tools um
2:06:05
prior to joining but also some tools um
2:06:05
prior to joining but also some tools um that i had personally used in a tutorial
2:06:08
that i had personally used in a tutorial
2:06:08
that i had personally used in a tutorial a couple of them
2:06:09
a couple of them
2:06:09
a couple of them for you guys to have a chance to go and
2:06:11
for you guys to have a chance to go and
2:06:11
for you guys to have a chance to go and build this yourself right we only had
2:06:13
build this yourself right we only had
2:06:13
build this yourself right we only had you know 30 25 30 minutes to talk here
2:06:16
you know 30 25 30 minutes to talk here
2:06:16
you know 30 25 30 minutes to talk here but the next step is doing it like get
2:06:19
but the next step is doing it like get
2:06:19
but the next step is doing it like get your fingers on a keyboard
2:06:20
your fingers on a keyboard
2:06:20
your fingers on a keyboard and build it you know you can use one of
2:06:22
and build it you know you can use one of
2:06:22
and build it you know you can use one of the models that are provided you can see
2:06:23
the models that are provided you can see
2:06:23
the models that are provided you can see down here
2:06:24
down here
2:06:24
down here leveraging fair learn but there's a lot
2:06:27
leveraging fair learn but there's a lot
2:06:27
leveraging fair learn but there's a lot to be said about the human
2:06:28
to be said about the human
2:06:28
to be said about the human discernment and the intention that we
2:06:30
discernment and the intention that we
2:06:30
discernment and the intention that we set as builders as
2:06:32
set as builders as
2:06:32
set as builders as modelers within these projects if you
2:06:35
modelers within these projects if you
2:06:35
modelers within these projects if you don't have
2:06:35
don't have
2:06:35
don't have i actually i hope that you have the
2:06:38
i actually i hope that you have the
2:06:38
i actually i hope that you have the support within your organization
2:06:41
support within your organization
2:06:41
support within your organization to be able to call a model bad when it's
2:06:43
to be able to call a model bad when it's
2:06:43
to be able to call a model bad when it's bad be able to pull the chain and say
2:06:46
bad be able to pull the chain and say
2:06:46
bad be able to pull the chain and say this has to stop we're doing the wrong
2:06:48
this has to stop we're doing the wrong
2:06:48
this has to stop we're doing the wrong thing but if you don't
2:06:50
thing but if you don't
2:06:50
thing but if you don't encourage your organizations to join you
2:06:53
encourage your organizations to join you
2:06:53
encourage your organizations to join you know organizations like
2:06:54
know organizations like
2:06:54
know organizations like partnerships on ai or mira or um
2:06:58
partnerships on ai or mira or um
2:06:58
partnerships on ai or mira or um rates or there's all these different
2:07:00
rates or there's all these different
2:07:00
rates or there's all these different organizations out there
2:07:01
organizations out there
2:07:01
organizations out there that will help companies align with
2:07:04
that will help companies align with
2:07:04
that will help companies align with ethical and responsible ai solutions and
2:07:06
ethical and responsible ai solutions and
2:07:06
ethical and responsible ai solutions and strategies
2:07:07
strategies
2:07:07
strategies that serve their bottom line that
2:07:09
that serve their bottom line that
2:07:09
that serve their bottom line that actually align with their business
2:07:11
actually align with their business
2:07:11
actually align with their business values as well
2:07:12
values as well
2:07:12
values as well so um i think that's probably all i have
2:07:15
so um i think that's probably all i have
2:07:15
so um i think that's probably all i have time for i did want to give you a chance
2:07:16
time for i did want to give you a chance
2:07:16
time for i did want to give you a chance to connect with me if you want to learn
2:07:18
to connect with me if you want to learn
2:07:18
to connect with me if you want to learn more
2:07:18
more
2:07:18
more i do have a a bunch of different
2:07:21
i do have a a bunch of different
2:07:21
i do have a a bunch of different tutorials and videos and resources that
2:07:22
tutorials and videos and resources that
2:07:22
tutorials and videos and resources that i always send to people so if you
2:07:24
i always send to people so if you
2:07:24
i always send to people so if you connect with me on linkedin i'll just
2:07:26
connect with me on linkedin i'll just
2:07:26
connect with me on linkedin i'll just copy and paste my resources right in
2:07:28
copy and paste my resources right in
2:07:28
copy and paste my resources right in there
2:07:28
there
2:07:28
there there's also a set in this deck that
2:07:30
there's also a set in this deck that
2:07:30
there's also a set in this deck that you'll get at the end but
2:07:31
you'll get at the end but
2:07:31
you'll get at the end but but it's just the beginning so thank you
2:07:33
but it's just the beginning so thank you
2:07:33
but it's just the beginning so thank you so much for having me
2:07:34
so much for having me
2:07:34
so much for having me um it has been amazing to be part of
2:07:37
um it has been amazing to be part of
2:07:37
um it has been amazing to be part of this full day
2:07:38
this full day
2:07:38
this full day uh and just think ethics right we're
2:07:41
uh and just think ethics right we're
2:07:41
uh and just think ethics right we're here we have a lot of power
2:07:42
here we have a lot of power
2:07:42
here we have a lot of power let's be responsible with it thank you
2:07:49
great thank you and the crowd goes wild
2:07:54
great thank you and the crowd goes wild
2:07:54
great thank you and the crowd goes wild [Laughter]
2:08:00
we do have some time with you in our
2:08:03
we do have some time with you in our
2:08:03
we do have some time with you in our panel discussion
2:08:04
panel discussion
2:08:04
panel discussion uh but first up next we have seth
2:08:07
uh but first up next we have seth
2:08:07
uh but first up next we have seth florence joining us
2:08:09
florence joining us
2:08:09
florence joining us okay bye bye
2:08:13
oh glad to have you back sir
2:08:16
oh glad to have you back sir
2:08:16
oh glad to have you back sir can you all hear me yeah loud and clear
2:08:21
can you all hear me yeah loud and clear
2:08:21
can you all hear me yeah loud and clear it's working yes how's everybody doing
2:08:23
it's working yes how's everybody doing
2:08:23
it's working yes how's everybody doing today they're
2:08:24
today they're
2:08:24
today they're pretty good really fun
2:08:28
fun fact fun fact alicia and i
2:08:31
fun fact fun fact alicia and i
2:08:31
fun fact fun fact alicia and i went to school together really yes
2:08:35
went to school together really yes
2:08:35
went to school together really yes graduated this same year from the same
2:08:37
graduated this same year from the same
2:08:37
graduated this same year from the same college
2:08:38
college
2:08:38
college and she was out shining us all
2:08:41
and she was out shining us all
2:08:41
and she was out shining us all that was what five years ago right yeah
2:08:44
that was what five years ago right yeah
2:08:44
that was what five years ago right yeah that's right
2:08:45
that's right
2:08:45
that's right look five years ago
2:08:49
look five years ago
2:08:49
look five years ago that's right so how's everybody doing uh
2:08:52
that's right so how's everybody doing uh
2:08:52
that's right so how's everybody doing uh we uh
2:08:52
we uh
2:08:52
we uh having a good time yeah we're we're
2:08:55
having a good time yeah we're we're
2:08:55
having a good time yeah we're we're having an
2:08:55
having an
2:08:55
having an excellent time we discuss drones uh
2:08:59
excellent time we discuss drones uh
2:08:59
excellent time we discuss drones uh self-flying drones even that are capable
2:09:01
self-flying drones even that are capable
2:09:02
self-flying drones even that are capable of detecting people in the water
2:09:03
of detecting people in the water
2:09:03
of detecting people in the water so there should be some deep learning in
2:09:05
so there should be some deep learning in
2:09:05
so there should be some deep learning in there by the way
2:09:07
there by the way
2:09:07
there by the way and we've had a talk about uh
2:09:09
and we've had a talk about uh
2:09:09
and we've had a talk about uh explainability
2:09:10
explainability
2:09:10
explainability and how that that's important i've been
2:09:13
and how that that's important i've been
2:09:13
and how that that's important i've been watching it's like this is the good
2:09:15
watching it's like this is the good
2:09:15
watching it's like this is the good programming for today and so it's all
2:09:17
programming for today and so it's all
2:09:17
programming for today and so it's all good stuff
2:09:18
good stuff
2:09:18
good stuff yeah we're getting into the advanced
2:09:20
yeah we're getting into the advanced
2:09:20
yeah we're getting into the advanced stuff and
2:09:21
stuff and
2:09:21
stuff and yeah so the thing that i'm interested in
2:09:24
yeah so the thing that i'm interested in
2:09:24
yeah so the thing that i'm interested in and that we're going to be talking about
2:09:26
and that we're going to be talking about
2:09:26
and that we're going to be talking about now is what is actually going on
2:09:29
now is what is actually going on
2:09:29
now is what is actually going on in a deep learning model and how it
2:09:31
in a deep learning model and how it
2:09:31
in a deep learning model and how it actually
2:09:33
actually
2:09:33
actually looks at stuff so we're going to get
2:09:34
looks at stuff so we're going to get
2:09:34
looks at stuff so we're going to get into the actual
2:09:36
into the actual
2:09:36
into the actual nitty-gritty of all of it and if my hair
2:09:38
nitty-gritty of all of it and if my hair
2:09:38
nitty-gritty of all of it and if my hair gets all crazy it's because when i talk
2:09:40
gets all crazy it's because when i talk
2:09:40
gets all crazy it's because when i talk about maths
2:09:41
about maths
2:09:41
about maths i mess up my hair because then everyone
2:09:43
i mess up my hair because then everyone
2:09:43
i mess up my hair because then everyone thinks it's smarter that way
2:09:45
thinks it's smarter that way
2:09:45
thinks it's smarter that way i pulled out all my hair by now i mean
2:09:48
i pulled out all my hair by now i mean
2:09:48
i pulled out all my hair by now i mean i've got
2:09:49
i've got
2:09:49
i've got almost nothing left because of all the
2:09:51
almost nothing left because of all the
2:09:51
almost nothing left because of all the maths involved in deep learning and
2:09:54
maths involved in deep learning and
2:09:54
maths involved in deep learning and and machine learning in general so i'm
2:09:56
and machine learning in general so i'm
2:09:56
and machine learning in general so i'm curious
2:09:57
curious
2:09:57
curious can you show us i am going to
2:10:02
please share my screen if that's okay
2:10:05
please share my screen if that's okay
2:10:05
please share my screen if that's okay it's absolutely so um basically what
2:10:07
it's absolutely so um basically what
2:10:08
it's absolutely so um basically what we're going to do
2:10:09
we're going to do
2:10:09
we're going to do is i am going to uh show you here
2:10:12
is i am going to uh show you here
2:10:12
is i am going to uh show you here let me minimize this thing and by the
2:10:13
let me minimize this thing and by the
2:10:13
let me minimize this thing and by the way if you have any questions please
2:10:15
way if you have any questions please
2:10:15
way if you have any questions please please uh ask them because i'm super
2:10:17
please uh ask them because i'm super
2:10:18
please uh ask them because i'm super interested and i'm gonna go like
2:10:19
interested and i'm gonna go like
2:10:19
interested and i'm gonna go like really really fast through a bunch of
2:10:22
really really fast through a bunch of
2:10:22
really really fast through a bunch of concepts
2:10:23
concepts
2:10:23
concepts uh and some maths so basically what
2:10:25
uh and some maths so basically what
2:10:26
uh and some maths so basically what we're doing
2:10:26
we're doing
2:10:26
we're doing right now is i'm going to show you how
2:10:28
right now is i'm going to show you how
2:10:28
right now is i'm going to show you how to build
2:10:29
to build
2:10:29
to build you saw the cognitive service early on
2:10:31
you saw the cognitive service early on
2:10:31
you saw the cognitive service early on from anthony
2:10:32
from anthony
2:10:32
from anthony right and then you saw noelle expertly
2:10:35
right and then you saw noelle expertly
2:10:35
right and then you saw noelle expertly take a look at
2:10:36
take a look at
2:10:36
take a look at how to make them explainable i am
2:10:38
how to make them explainable i am
2:10:38
how to make them explainable i am actually going to explain what is
2:10:39
actually going to explain what is
2:10:39
actually going to explain what is happening
2:10:39
happening
2:10:40
happening inside of these models okay and so what
2:10:42
inside of these models okay and so what
2:10:42
inside of these models okay and so what we're building effectively is this
2:10:44
we're building effectively is this
2:10:44
we're building effectively is this we are building an ai uh or a machine
2:10:47
we are building an ai uh or a machine
2:10:47
we are building an ai uh or a machine learning model that
2:10:48
learning model that
2:10:48
learning model that looks at pictures and tells you whether
2:10:50
looks at pictures and tells you whether
2:10:50
looks at pictures and tells you whether it's a taco
2:10:52
it's a taco
2:10:52
it's a taco or a burrito this is more of an
2:10:53
or a burrito this is more of an
2:10:53
or a burrito this is more of an enchilada but still
2:10:55
enchilada but still
2:10:55
enchilada but still if it's a burrito i don't know how that
2:10:57
if it's a burrito i don't know how that
2:10:57
if it's a burrito i don't know how that one got in there but if it's a burrito
2:10:59
one got in there but if it's a burrito
2:10:59
one got in there but if it's a burrito or a taco because this is what we should
2:11:02
or a taco because this is what we should
2:11:02
or a taco because this is what we should be looking at when it comes
2:11:04
be looking at when it comes
2:11:04
be looking at when it comes to ai we need to be looking at whether
2:11:06
to ai we need to be looking at whether
2:11:06
to ai we need to be looking at whether foods are good or bad
2:11:07
foods are good or bad
2:11:07
foods are good or bad right and burritos versus tacos is the
2:11:10
right and burritos versus tacos is the
2:11:10
right and burritos versus tacos is the quintessential
2:11:11
quintessential
2:11:11
quintessential machine learning project and what i'm
2:11:12
machine learning project and what i'm
2:11:12
machine learning project and what i'm going to show you is basically i built a
2:11:15
going to show you is basically i built a
2:11:15
going to show you is basically i built a machine learning model
2:11:16
machine learning model
2:11:16
machine learning model that actually looks at pictures of these
2:11:19
that actually looks at pictures of these
2:11:19
that actually looks at pictures of these things and tells you
2:11:20
things and tells you
2:11:20
things and tells you whether they're burritos or tacos right
2:11:22
whether they're burritos or tacos right
2:11:22
whether they're burritos or tacos right so i already ran it
2:11:24
so i already ran it
2:11:24
so i already ran it sometimes my gpu doesn't like it if i'm
2:11:26
sometimes my gpu doesn't like it if i'm
2:11:26
sometimes my gpu doesn't like it if i'm doing too many things with the gpu
2:11:28
doing too many things with the gpu
2:11:28
doing too many things with the gpu and so if it doesn't work i've already
2:11:29
and so if it doesn't work i've already
2:11:29
and so if it doesn't work i've already ran these things uh but there you go
2:11:32
ran these things uh but there you go
2:11:32
ran these things uh but there you go so basically you can see here that this
2:11:34
so basically you can see here that this
2:11:34
so basically you can see here that this picture it says
2:11:35
picture it says
2:11:35
picture it says it's a burrito right it even tells you
2:11:38
it's a burrito right it even tells you
2:11:38
it's a burrito right it even tells you the scores and this picture says
2:11:40
the scores and this picture says
2:11:40
the scores and this picture says it's a taco uh taco by the way and you
2:11:43
it's a taco uh taco by the way and you
2:11:43
it's a taco uh taco by the way and you can see the score
2:11:44
can see the score
2:11:44
can see the score as well and if you want i can go to this
2:11:46
as well and if you want i can go to this
2:11:46
as well and if you want i can go to this particular picture here
2:11:48
particular picture here
2:11:48
particular picture here and let's go to el browser uh the
2:11:51
and let's go to el browser uh the
2:11:51
and let's go to el browser uh the browser here
2:11:52
browser here
2:11:52
browser here and let me bring it over let me put the
2:11:54
and let me bring it over let me put the
2:11:54
and let me bring it over let me put the picture in there so you can see it
2:11:56
picture in there so you can see it
2:11:56
picture in there so you can see it oh my gosh i gave it a pizza and it
2:11:58
oh my gosh i gave it a pizza and it
2:11:58
oh my gosh i gave it a pizza and it thought it was a burrito there's the
2:12:00
thought it was a burrito there's the
2:12:00
thought it was a burrito there's the joke jokes on me but
2:12:01
joke jokes on me but
2:12:01
joke jokes on me but you can see that as soon as we get into
2:12:03
you can see that as soon as we get into
2:12:03
you can see that as soon as we get into the ethics of this
2:12:04
the ethics of this
2:12:04
the ethics of this you're realizing that when it looks at a
2:12:06
you're realizing that when it looks at a
2:12:06
you're realizing that when it looks at a pizza
2:12:09
pizza
2:12:09
pizza it thinks it's a burrito and you're
2:12:12
it thinks it's a burrito and you're
2:12:12
it thinks it's a burrito and you're probably wondering
2:12:13
probably wondering
2:12:13
probably wondering why and i'm going to show you what and
2:12:17
why and i'm going to show you what and
2:12:17
why and i'm going to show you what and and this and i'm going to show you to a
2:12:18
and this and i'm going to show you to a
2:12:18
and this and i'm going to show you to a degree where you're going to be
2:12:20
degree where you're going to be
2:12:20
degree where you're going to be you might not understand the whole
2:12:21
you might not understand the whole
2:12:21
you might not understand the whole infrastructure but you're going to
2:12:23
infrastructure but you're going to
2:12:23
infrastructure but you're going to understand
2:12:23
understand
2:12:24
understand fundamentally what it is that this thing
2:12:27
fundamentally what it is that this thing
2:12:27
fundamentally what it is that this thing is doing
2:12:28
is doing
2:12:28
is doing and then by extension you're going to
2:12:31
and then by extension you're going to
2:12:31
and then by extension you're going to immediately understand
2:12:33
immediately understand
2:12:33
immediately understand why we need to understand how it works
2:12:35
why we need to understand how it works
2:12:35
why we need to understand how it works in order to make ethical
2:12:36
in order to make ethical
2:12:36
in order to make ethical choices so basically when we're talking
2:12:39
choices so basically when we're talking
2:12:39
choices so basically when we're talking about machine
2:12:41
about machine
2:12:41
about machine learning models model is for me is a
2:12:43
learning models model is for me is a
2:12:44
learning models model is for me is a lazy way of writing a function
2:12:45
lazy way of writing a function
2:12:45
lazy way of writing a function with data just think whenever if you're
2:12:47
with data just think whenever if you're
2:12:47
with data just think whenever if you're a programmer and you think of the word
2:12:49
a programmer and you think of the word
2:12:49
a programmer and you think of the word model just think function
2:12:50
model just think function
2:12:50
model just think function and a function is this function thing
2:12:53
and a function is this function thing
2:12:53
and a function is this function thing but the model we have to tell it what
2:12:56
but the model we have to tell it what
2:12:56
but the model we have to tell it what the shape of the function looks like
2:12:58
the shape of the function looks like
2:12:58
the shape of the function looks like and then there's internal parameters
2:13:00
and then there's internal parameters
2:13:00
and then there's internal parameters that it learns in our case with deep
2:13:02
that it learns in our case with deep
2:13:02
that it learns in our case with deep learning
2:13:03
learning
2:13:03
learning it's just numbers right so you got to
2:13:05
it's just numbers right so you got to
2:13:05
it's just numbers right so you got to first think about this model
2:13:07
first think about this model
2:13:07
first think about this model uh it's a function right then you need
2:13:09
uh it's a function right then you need
2:13:09
uh it's a function right then you need to think about this notion of a cost or
2:13:11
to think about this notion of a cost or
2:13:11
to think about this notion of a cost or a loss function this tells us how bad we
2:13:13
a loss function this tells us how bad we
2:13:13
a loss function this tells us how bad we are at it so
2:13:14
are at it so
2:13:14
are at it so again let me i'm going to be super fancy
2:13:16
again let me i'm going to be super fancy
2:13:16
again let me i'm going to be super fancy and use the pen here
2:13:17
and use the pen here
2:13:17
and use the pen here this thing right here is the model that
2:13:19
this thing right here is the model that
2:13:20
this thing right here is the model that we learned
2:13:21
we learned
2:13:21
we learned this here is the actual right answer and
2:13:24
this here is the actual right answer and
2:13:24
this here is the actual right answer and this function tells us
2:13:25
this function tells us
2:13:25
this function tells us how sucky we are at it so for example
2:13:29
how sucky we are at it so for example
2:13:29
how sucky we are at it so for example if the loss is zero that means that the
2:13:32
if the loss is zero that means that the
2:13:32
if the loss is zero that means that the difference between these two things
2:13:34
difference between these two things
2:13:34
difference between these two things right if you were to subtract them is
2:13:36
right if you were to subtract them is
2:13:36
right if you were to subtract them is zero which means it's exactly right
2:13:38
zero which means it's exactly right
2:13:38
zero which means it's exactly right whenever these things are not right you
2:13:40
whenever these things are not right you
2:13:40
whenever these things are not right you get an answer that's
2:13:41
get an answer that's
2:13:41
get an answer that's non-zero from this cost function and the
2:13:44
non-zero from this cost function and the
2:13:44
non-zero from this cost function and the optimizer's job
2:13:46
optimizer's job
2:13:46
optimizer's job is to minimize the loss
2:13:49
is to minimize the loss
2:13:50
is to minimize the loss that we get from our model and our
2:13:52
that we get from our model and our
2:13:52
that we get from our model and our actual
2:13:53
actual
2:13:53
actual target which is the actual answer so in
2:13:55
target which is the actual answer so in
2:13:55
target which is the actual answer so in this case we're gonna have a bunch of
2:13:57
this case we're gonna have a bunch of
2:13:57
this case we're gonna have a bunch of pictures
2:13:58
pictures
2:13:58
pictures uh that's the x and then we're gonna
2:14:00
uh that's the x and then we're gonna
2:14:00
uh that's the x and then we're gonna have a bunch of
2:14:01
have a bunch of
2:14:01
have a bunch of labels that says that's a taco that's a
2:14:03
labels that says that's a taco that's a
2:14:03
labels that says that's a taco that's a burrito that's a taco that's a burrito
2:14:05
burrito that's a taco that's a burrito
2:14:05
burrito that's a taco that's a burrito but it's important to recognize how we
2:14:07
but it's important to recognize how we
2:14:07
but it's important to recognize how we set up the problem so that when we give
2:14:09
set up the problem so that when we give
2:14:09
set up the problem so that when we give it this thing
2:14:10
it this thing
2:14:10
it this thing the pizza we know why it's saying it's a
2:14:13
the pizza we know why it's saying it's a
2:14:13
the pizza we know why it's saying it's a burrito
2:14:14
burrito
2:14:14
burrito with such high confidence versus a taco
2:14:16
with such high confidence versus a taco
2:14:16
with such high confidence versus a taco okay
2:14:17
okay
2:14:17
okay all right so by the way uh get your
2:14:19
all right so by the way uh get your
2:14:19
all right so by the way uh get your questions in and just interrupt me as we
2:14:20
questions in and just interrupt me as we
2:14:20
questions in and just interrupt me as we go i don't have a lot of time
2:14:22
go i don't have a lot of time
2:14:22
go i don't have a lot of time and i'm trying to cover like a couple of
2:14:25
and i'm trying to cover like a couple of
2:14:25
and i'm trying to cover like a couple of um
2:14:26
um
2:14:26
um semesters of grad school class so let's
2:14:27
semesters of grad school class so let's
2:14:28
semesters of grad school class so let's talk about creating a model
2:14:29
talk about creating a model
2:14:29
talk about creating a model so how do we make this h thing right and
2:14:31
so how do we make this h thing right and
2:14:32
so how do we make this h thing right and i'm gonna we're gonna build a little
2:14:33
i'm gonna we're gonna build a little
2:14:33
i'm gonna we're gonna build a little a really tiny dumb one and then we're
2:14:35
a really tiny dumb one and then we're
2:14:35
a really tiny dumb one and then we're gonna construct it all up into exactly
2:14:37
gonna construct it all up into exactly
2:14:37
gonna construct it all up into exactly what it is that we're doing
2:14:39
what it is that we're doing
2:14:39
what it is that we're doing uh with um the burritos and tacos and if
2:14:41
uh with um the burritos and tacos and if
2:14:41
uh with um the burritos and tacos and if you're wondering we're using inception
2:14:42
you're wondering we're using inception
2:14:42
you're wondering we're using inception v2
2:14:43
v2
2:14:43
v2 on that one with transfer learning so
2:14:45
on that one with transfer learning so
2:14:45
on that one with transfer learning so first what does the x look like and what
2:14:47
first what does the x look like and what
2:14:47
first what does the x look like and what i'm gonna do is i'm gonna shrink the
2:14:48
i'm gonna do is i'm gonna shrink the
2:14:48
i'm gonna do is i'm gonna shrink the problem down to something really dumb
2:14:50
problem down to something really dumb
2:14:50
problem down to something really dumb we're going to make a machine learning
2:14:52
we're going to make a machine learning
2:14:52
we're going to make a machine learning model to look at a 9
2:14:54
model to look at a 9
2:14:54
model to look at a 9 pixel picture a grayscale picture and
2:14:57
pixel picture a grayscale picture and
2:14:57
pixel picture a grayscale picture and we're going to guess whether it's darker
2:14:58
we're going to guess whether it's darker
2:14:58
we're going to guess whether it's darker at the top or darker at the bottom
2:15:00
at the top or darker at the bottom
2:15:00
at the top or darker at the bottom so we got to build a construct so i'm
2:15:02
so we got to build a construct so i'm
2:15:02
so we got to build a construct so i'm just going to build one we're going to
2:15:03
just going to build one we're going to
2:15:03
just going to build one we're going to build this thing right here remember
2:15:04
build this thing right here remember
2:15:04
build this thing right here remember there's 9 pixels
2:15:06
there's 9 pixels
2:15:06
there's 9 pixels x1 through x9 those values range between
2:15:09
x1 through x9 those values range between
2:15:09
x1 through x9 those values range between 0 and 255
2:15:11
0 and 255
2:15:11
0 and 255 because it's a grayscale image and your
2:15:13
because it's a grayscale image and your
2:15:13
because it's a grayscale image and your job is to invent
2:15:14
job is to invent
2:15:14
job is to invent w's such that when you multiply all
2:15:17
w's such that when you multiply all
2:15:17
w's such that when you multiply all those pixels together
2:15:20
those pixels together
2:15:20
those pixels together you get an answer whether it's darker at
2:15:21
you get an answer whether it's darker at
2:15:21
you get an answer whether it's darker at the top or darker at the bottom let me
2:15:23
the top or darker at the bottom let me
2:15:23
the top or darker at the bottom let me go back because i went really fast
2:15:25
go back because i went really fast
2:15:25
go back because i went really fast here's the picture uh if you look right
2:15:27
here's the picture uh if you look right
2:15:27
here's the picture uh if you look right here here are the numbers
2:15:28
here here are the numbers
2:15:28
here here are the numbers notice that those are the top three
2:15:30
notice that those are the top three
2:15:30
notice that those are the top three numbers here right and then these are
2:15:32
numbers here right and then these are
2:15:32
numbers here right and then these are the bottom three numbers here
2:15:33
the bottom three numbers here
2:15:33
the bottom three numbers here right right here so these are these
2:15:35
right right here so these are these
2:15:35
right right here so these are these pictures so for this one it's clearly
2:15:37
pictures so for this one it's clearly
2:15:37
pictures so for this one it's clearly darker at the bottom
2:15:38
darker at the bottom
2:15:38
darker at the bottom so you can see that if these numbers
2:15:41
so you can see that if these numbers
2:15:41
so you can see that if these numbers here
2:15:42
here
2:15:42
here together versus these numbers here
2:15:44
together versus these numbers here
2:15:44
together versus these numbers here together
2:15:45
together
2:15:45
together that will tell you what you've got to do
2:15:46
that will tell you what you've got to do
2:15:46
that will tell you what you've got to do but i forced you to construct it this
2:15:48
but i forced you to construct it this
2:15:48
but i forced you to construct it this way okay so let me show you how it works
2:15:51
way okay so let me show you how it works
2:15:51
way okay so let me show you how it works so here is a picture that's clearly
2:15:53
so here is a picture that's clearly
2:15:53
so here is a picture that's clearly darker at the top
2:15:54
darker at the top
2:15:54
darker at the top whenever i put these numbers in i guess
2:15:57
whenever i put these numbers in i guess
2:15:57
whenever i put these numbers in i guess these numbers notice that the answer is
2:15:59
these numbers notice that the answer is
2:15:59
these numbers notice that the answer is going to be
2:16:00
going to be
2:16:00
going to be positive which tells us it's darker at
2:16:02
positive which tells us it's darker at
2:16:02
positive which tells us it's darker at the top okay let's go to the
2:16:04
the top okay let's go to the
2:16:04
the top okay let's go to the one at the bottom i'm going to add these
2:16:05
one at the bottom i'm going to add these
2:16:05
one at the bottom i'm going to add these numbers so i'm i'm
2:16:07
numbers so i'm i'm
2:16:07
numbers so i'm i'm hand crafting these fake w's such that
2:16:10
hand crafting these fake w's such that
2:16:10
hand crafting these fake w's such that when i pass these numbers in notice that
2:16:11
when i pass these numbers in notice that
2:16:11
when i pass these numbers in notice that when i do this
2:16:12
when i do this
2:16:12
when i do this we get a negative answer so what is
2:16:13
we get a negative answer so what is
2:16:14
we get a negative answer so what is actually happening these numbers are
2:16:15
actually happening these numbers are
2:16:15
actually happening these numbers are multiplied by these numbers
2:16:17
multiplied by these numbers
2:16:17
multiplied by these numbers like so right and then once those
2:16:19
like so right and then once those
2:16:19
like so right and then once those numbers are multiplied
2:16:20
numbers are multiplied
2:16:20
numbers are multiplied we're adding them all together and then
2:16:22
we're adding them all together and then
2:16:22
we're adding them all together and then out comes an answer which is this thing
2:16:23
out comes an answer which is this thing
2:16:23
out comes an answer which is this thing right here
2:16:24
right here
2:16:24
right here notice that pretty easily you're able to
2:16:27
notice that pretty easily you're able to
2:16:27
notice that pretty easily you're able to create
2:16:28
create
2:16:28
create w's yourself such that you can
2:16:30
w's yourself such that you can
2:16:30
w's yourself such that you can guesstimate or
2:16:31
guesstimate or
2:16:32
guesstimate or classify this rudimentary picture into
2:16:34
classify this rudimentary picture into
2:16:34
classify this rudimentary picture into darkness at the bottom versus darkness
2:16:36
darkness at the bottom versus darkness
2:16:36
darkness at the bottom versus darkness at the top
2:16:36
at the top
2:16:36
at the top it's negative dark to the bottom
2:16:38
it's negative dark to the bottom
2:16:38
it's negative dark to the bottom positive darkness at the top
2:16:40
positive darkness at the top
2:16:40
positive darkness at the top now here is a goofy question that you're
2:16:42
now here is a goofy question that you're
2:16:42
now here is a goofy question that you're all happening because there's just
2:16:44
all happening because there's just
2:16:44
all happening because there's just there's this b where's my mouse oh i'm
2:16:46
there's this b where's my mouse oh i'm
2:16:46
there's this b where's my mouse oh i'm old enough to where i can't there it is
2:16:47
old enough to where i can't there it is
2:16:47
old enough to where i can't there it is so there's this b
2:16:48
so there's this b
2:16:48
so there's this b just hanging out here you're probably
2:16:50
just hanging out here you're probably
2:16:50
just hanging out here you're probably wondering what is this beat
2:16:51
wondering what is this beat
2:16:51
wondering what is this beat well here's the thing what if the
2:16:53
well here's the thing what if the
2:16:53
well here's the thing what if the problem is such that we want the top
2:16:55
problem is such that we want the top
2:16:56
problem is such that we want the top to be uh let's just say we want to guess
2:16:58
to be uh let's just say we want to guess
2:16:58
to be uh let's just say we want to guess that
2:16:59
that
2:16:59
that the top is if it's a little bit like if
2:17:01
the top is if it's a little bit like if
2:17:01
the top is if it's a little bit like if it's a certain percentage of darkness
2:17:03
it's a certain percentage of darkness
2:17:03
it's a certain percentage of darkness from the bottom versus the top we still
2:17:05
from the bottom versus the top we still
2:17:05
from the bottom versus the top we still want to guess top even though it's not
2:17:06
want to guess top even though it's not
2:17:06
want to guess top even though it's not darker
2:17:07
darker
2:17:07
darker we can add a positive number here such
2:17:09
we can add a positive number here such
2:17:09
we can add a positive number here such that it biases it towards answering top
2:17:11
that it biases it towards answering top
2:17:11
that it biases it towards answering top we can add a negative number here such
2:17:13
we can add a negative number here such
2:17:13
we can add a negative number here such as that it biases it to answer
2:17:15
as that it biases it to answer
2:17:15
as that it biases it to answer bottom in machine learning this
2:17:17
bottom in machine learning this
2:17:17
bottom in machine learning this particular term is called
2:17:19
particular term is called
2:17:20
particular term is called get ready the bias right so here what
2:17:23
get ready the bias right so here what
2:17:23
get ready the bias right so here what we've done is we've basically
2:17:24
we've done is we've basically
2:17:24
we've done is we've basically constructed
2:17:25
constructed
2:17:25
constructed a machine learning model right this
2:17:28
a machine learning model right this
2:17:28
a machine learning model right this w that we just formulated given this
2:17:32
w that we just formulated given this
2:17:32
w that we just formulated given this pixel is that h of x right we've
2:17:35
pixel is that h of x right we've
2:17:35
pixel is that h of x right we've basically constructed one
2:17:36
basically constructed one
2:17:36
basically constructed one such that we have to figure out what
2:17:38
such that we have to figure out what
2:17:38
such that we have to figure out what these parameters are and if the number
2:17:40
these parameters are and if the number
2:17:40
these parameters are and if the number is positive we guess top if the number
2:17:42
is positive we guess top if the number
2:17:42
is positive we guess top if the number is
2:17:42
is
2:17:42
is negative we have we guess bottom so this
2:17:45
negative we have we guess bottom so this
2:17:45
negative we have we guess bottom so this w and this b
2:17:46
w and this b
2:17:46
w and this b becomes the actual model construct and
2:17:48
becomes the actual model construct and
2:17:48
becomes the actual model construct and this becomes the prediction crazy right
2:17:50
this becomes the prediction crazy right
2:17:50
this becomes the prediction crazy right you would never use machine learning for
2:17:52
you would never use machine learning for
2:17:52
you would never use machine learning for this but i want you to see what's going
2:17:54
this but i want you to see what's going
2:17:54
this but i want you to see what's going on you might be thinking well seth what
2:17:55
on you might be thinking well seth what
2:17:55
on you might be thinking well seth what if we want to guess
2:17:56
if we want to guess
2:17:56
if we want to guess top middle or bottom well you just make
2:17:58
top middle or bottom well you just make
2:17:58
top middle or bottom well you just make three of them and now you're getting
2:17:59
three of them and now you're getting
2:17:59
three of them and now you're getting into things like linear algebra
2:18:01
into things like linear algebra
2:18:01
into things like linear algebra right where you take this uh this thing
2:18:04
right where you take this uh this thing
2:18:04
right where you take this uh this thing right here and you your vector on the
2:18:05
right here and you your vector on the
2:18:05
right here and you your vector on the side and you
2:18:05
side and you
2:18:06
side and you turn it on its side and go num nom nom
2:18:07
turn it on its side and go num nom nom
2:18:07
turn it on its side and go num nom nom and notice that what this is doing if we
2:18:09
and notice that what this is doing if we
2:18:09
and notice that what this is doing if we have three of them at the same time is
2:18:11
have three of them at the same time is
2:18:11
have three of them at the same time is this is adding the first three terms
2:18:13
this is adding the first three terms
2:18:13
this is adding the first three terms putting it in this one adding the second
2:18:15
putting it in this one adding the second
2:18:15
putting it in this one adding the second three putting them in here and then so
2:18:17
three putting them in here and then so
2:18:17
three putting them in here and then so on and so forth and we just guess the
2:18:18
on and so forth and we just guess the
2:18:18
on and so forth and we just guess the one that's
2:18:19
one that's
2:18:19
one that's darkest right we might even divide by
2:18:21
darkest right we might even divide by
2:18:22
darkest right we might even divide by the sum of this stuff
2:18:24
the sum of this stuff
2:18:24
the sum of this stuff and then this comes out looking like a
2:18:25
and then this comes out looking like a
2:18:25
and then this comes out looking like a percentage of some kind like 87
2:18:28
percentage of some kind like 87
2:18:28
percentage of some kind like 87 versus 62 percent versus whatever these
2:18:30
versus 62 percent versus whatever these
2:18:30
versus 62 percent versus whatever these need to add to one so this is completely
2:18:32
need to add to one so this is completely
2:18:32
need to add to one so this is completely wrong
2:18:33
wrong
2:18:33
wrong but you you get what i'm saying these
2:18:34
but you you get what i'm saying these
2:18:34
but you you get what i'm saying these are just sort of
2:18:36
are just sort of
2:18:36
are just sort of densities of the answer that are coming
2:18:38
densities of the answer that are coming
2:18:38
densities of the answer that are coming out that's why i'm very reluctant
2:18:40
out that's why i'm very reluctant
2:18:40
out that's why i'm very reluctant friends when you see percentages coming
2:18:42
friends when you see percentages coming
2:18:42
friends when you see percentages coming out
2:18:43
out
2:18:43
out of uh neural networks and you're saying
2:18:45
of uh neural networks and you're saying
2:18:46
of uh neural networks and you're saying oh that's the percent confidence
2:18:48
oh that's the percent confidence
2:18:48
oh that's the percent confidence i guess it can be that but it's not
2:18:50
i guess it can be that but it's not
2:18:50
i guess it can be that but it's not actual statistics
2:18:52
actual statistics
2:18:52
actual statistics so just be aware of that okay so again
2:18:54
so just be aware of that okay so again
2:18:54
so just be aware of that okay so again top middle bottom
2:18:55
top middle bottom
2:18:55
top middle bottom now instead of a single vector of w's
2:18:58
now instead of a single vector of w's
2:18:58
now instead of a single vector of w's and a single
2:18:59
and a single
2:18:59
and a single scalar of uh for the bias we have a
2:19:02
scalar of uh for the bias we have a
2:19:02
scalar of uh for the bias we have a matrix w for um uh
2:19:06
matrix w for um uh
2:19:06
matrix w for um uh for the weights and then we have a
2:19:07
for the weights and then we have a
2:19:07
for the weights and then we have a vector b for the biases this is what
2:19:09
vector b for the biases this is what
2:19:09
vector b for the biases this is what this
2:19:09
this
2:19:09
this looks like right if you were to put this
2:19:11
looks like right if you were to put this
2:19:11
looks like right if you were to put this in neural network
2:19:12
in neural network
2:19:12
in neural network drawing right all these lines the
2:19:14
drawing right all these lines the
2:19:14
drawing right all these lines the numbers that we're multiplying these
2:19:16
numbers that we're multiplying these
2:19:16
numbers that we're multiplying these things by
2:19:17
things by
2:19:17
things by the number 1 1 1 and then zero zero zero
2:19:19
the number 1 1 1 and then zero zero zero
2:19:19
the number 1 1 1 and then zero zero zero zero right
2:19:20
zero right
2:19:20
zero right and then out comes this number and then
2:19:21
and then out comes this number and then
2:19:21
and then out comes this number and then this is the density that we saw before
2:19:23
this is the density that we saw before
2:19:23
this is the density that we saw before for us
2:19:24
for us
2:19:24
for us there was three numbers right uh for
2:19:26
there was three numbers right uh for
2:19:26
there was three numbers right uh for larger problems like tacos or burritos
2:19:29
larger problems like tacos or burritos
2:19:29
larger problems like tacos or burritos the answer
2:19:29
the answer
2:19:29
the answer is gonna be of size two right uh but
2:19:34
is gonna be of size two right uh but
2:19:34
is gonna be of size two right uh but this really doesn't work for a bigger
2:19:36
this really doesn't work for a bigger
2:19:36
this really doesn't work for a bigger model so what if we make it bigger
2:19:38
model so what if we make it bigger
2:19:38
model so what if we make it bigger right what if we do this then all of a
2:19:40
right what if we do this then all of a
2:19:40
right what if we do this then all of a sudden you're thinking well now we're
2:19:41
sudden you're thinking well now we're
2:19:41
sudden you're thinking well now we're making a neural network right and you're
2:19:43
making a neural network right and you're
2:19:43
making a neural network right and you're like oh wow this is starting to get into
2:19:45
like oh wow this is starting to get into
2:19:45
like oh wow this is starting to get into to coolness territory well it turns out
2:19:47
to coolness territory well it turns out
2:19:47
to coolness territory well it turns out that if you were to just stack them and
2:19:49
that if you were to just stack them and
2:19:49
that if you were to just stack them and then do these matrix multiplications
2:19:51
then do these matrix multiplications
2:19:51
then do these matrix multiplications the problem with uh linear models is
2:19:53
the problem with uh linear models is
2:19:53
the problem with uh linear models is that they force what's called a linear
2:19:55
that they force what's called a linear
2:19:55
that they force what's called a linear separation of the
2:19:57
separation of the
2:19:57
separation of the space that you're trying to guess so in
2:19:59
space that you're trying to guess so in
2:19:59
space that you're trying to guess so in your mind's eye think of points in 3d
2:20:01
your mind's eye think of points in 3d
2:20:01
your mind's eye think of points in 3d space
2:20:01
space
2:20:02
space right the w what it does is it makes a a
2:20:04
right the w what it does is it makes a a
2:20:04
right the w what it does is it makes a a plane in 3d space and
2:20:06
plane in 3d space and
2:20:06
plane in 3d space and separates the points for you but what if
2:20:08
separates the points for you but what if
2:20:08
separates the points for you but what if the points are not
2:20:09
the points are not
2:20:09
the points are not able to be separated by a line what if
2:20:12
able to be separated by a line what if
2:20:12
able to be separated by a line what if you needed to have some complex
2:20:13
you needed to have some complex
2:20:13
you needed to have some complex separation
2:20:14
separation
2:20:14
separation because effectively what you're doing is
2:20:15
because effectively what you're doing is
2:20:15
because effectively what you're doing is you're trying to come up with a function
2:20:17
you're trying to come up with a function
2:20:17
you're trying to come up with a function that
2:20:17
that
2:20:17
that separates the classes of things that
2:20:19
separates the classes of things that
2:20:19
separates the classes of things that you're trying to guess right
2:20:20
you're trying to guess right
2:20:20
you're trying to guess right so it turns out that just stacking them
2:20:22
so it turns out that just stacking them
2:20:22
so it turns out that just stacking them doesn't do anything
2:20:23
doesn't do anything
2:20:24
doesn't do anything because a linear combination which is
2:20:25
because a linear combination which is
2:20:25
because a linear combination which is what this is of a linear combination
2:20:28
what this is of a linear combination
2:20:28
what this is of a linear combination which is what this is
2:20:29
which is what this is
2:20:29
which is what this is is still a linear combination and you
2:20:30
is still a linear combination and you
2:20:30
is still a linear combination and you can prove that simply by using
2:20:32
can prove that simply by using
2:20:32
can prove that simply by using mathematical induction
2:20:34
mathematical induction
2:20:34
mathematical induction which is really cool right i'm not going
2:20:36
which is really cool right i'm not going
2:20:36
which is really cool right i'm not going to get into it because i have like 15
2:20:37
to get into it because i have like 15
2:20:37
to get into it because i have like 15 minutes to finish
2:20:38
minutes to finish
2:20:38
minutes to finish semester two of neural networks right uh
2:20:40
semester two of neural networks right uh
2:20:40
semester two of neural networks right uh in a grad school course
2:20:42
in a grad school course
2:20:42
in a grad school course so it turns out that you can keep
2:20:44
so it turns out that you can keep
2:20:44
so it turns out that you can keep stacking them like this
2:20:45
stacking them like this
2:20:45
stacking them like this but it won't make a liquid difference
2:20:47
but it won't make a liquid difference
2:20:47
but it won't make a liquid difference unless you introduce what's called a
2:20:49
unless you introduce what's called a
2:20:49
unless you introduce what's called a non-linearity
2:20:50
non-linearity
2:20:50
non-linearity and the way to do that is to add what's
2:20:51
and the way to do that is to add what's
2:20:52
and the way to do that is to add what's called an activation function in between
2:20:54
called an activation function in between
2:20:54
called an activation function in between so you're multiplying this matrix times
2:20:56
so you're multiplying this matrix times
2:20:56
so you're multiplying this matrix times that thing right then you're you're
2:20:58
that thing right then you're you're
2:20:58
that thing right then you're you're you're you're you're running a function
2:20:59
you're you're you're running a function
2:20:59
you're you're you're running a function across all of the things in there
2:21:01
across all of the things in there
2:21:01
across all of the things in there introduces and then you do it again and
2:21:03
introduces and then you do it again and
2:21:03
introduces and then you do it again and you do it again and then all of a sudden
2:21:04
you do it again and then all of a sudden
2:21:04
you do it again and then all of a sudden you start to have
2:21:05
you start to have
2:21:05
you start to have things like this which are neural
2:21:07
things like this which are neural
2:21:07
things like this which are neural networks right this neural network looks
2:21:09
networks right this neural network looks
2:21:09
networks right this neural network looks like it takes a picture of a
2:21:11
like it takes a picture of a
2:21:11
like it takes a picture of a of a number and then it it multiplies it
2:21:14
of a number and then it it multiplies it
2:21:14
of a number and then it it multiplies it right
2:21:14
right
2:21:14
right and then it runs the function over and
2:21:15
and then it runs the function over and
2:21:16
and then it runs the function over and then it multiplies again there and
2:21:17
then it multiplies again there and
2:21:17
then it multiplies again there and notice that as the network
2:21:18
notice that as the network
2:21:18
notice that as the network goes the space like shrinks down
2:21:22
goes the space like shrinks down
2:21:22
goes the space like shrinks down right so notice that at the end since
2:21:23
right so notice that at the end since
2:21:24
right so notice that at the end since this is the digits problem it's only
2:21:25
this is the digits problem it's only
2:21:25
this is the digits problem it's only going to have a density of 10 because
2:21:27
going to have a density of 10 because
2:21:27
going to have a density of 10 because right there's 10 spaces and if the if
2:21:29
right there's 10 spaces and if the if
2:21:29
right there's 10 spaces and if the if it's the highest
2:21:30
it's the highest
2:21:30
it's the highest number isn't here then we get zero if
2:21:32
number isn't here then we get zero if
2:21:32
number isn't here then we get zero if the highest number's in here then we get
2:21:33
the highest number's in here then we get
2:21:33
the highest number's in here then we get nine right
2:21:34
nine right
2:21:34
nine right and that's that's how this goes but it
2:21:36
and that's that's how this goes but it
2:21:36
and that's that's how this goes but it turns out that with images
2:21:38
turns out that with images
2:21:38
turns out that with images right if you were to take an image and
2:21:39
right if you were to take an image and
2:21:39
right if you were to take an image and then strip out the color and then
2:21:41
then strip out the color and then
2:21:41
then strip out the color and then line up everything in a single row
2:21:43
line up everything in a single row
2:21:44
line up everything in a single row you're going to miss
2:21:45
you're going to miss
2:21:45
you're going to miss things like for example pixels on top of
2:21:47
things like for example pixels on top of
2:21:47
things like for example pixels on top of each other and so there's these
2:21:49
each other and so there's these
2:21:49
each other and so there's these interesting
2:21:49
interesting
2:21:49
interesting neural networks called convolutional
2:21:51
neural networks called convolutional
2:21:51
neural networks called convolutional neural networks which do like a
2:21:53
neural networks which do like a
2:21:53
neural networks which do like a like a dot product over images right let
2:21:55
like a dot product over images right let
2:21:55
like a dot product over images right let me show you what that looks like
2:21:57
me show you what that looks like
2:21:57
me show you what that looks like so here is um here's visual studio code
2:22:01
so here is um here's visual studio code
2:22:01
so here is um here's visual studio code and what i've done
2:22:02
and what i've done
2:22:02
and what i've done already for you is i've made a
2:22:05
already for you is i've made a
2:22:05
already for you is i've made a convolution
2:22:05
convolution
2:22:06
convolution so let's go here and let me uh do this
2:22:08
so let's go here and let me uh do this
2:22:08
so let's go here and let me uh do this filter right here by the way doesn't
2:22:09
filter right here by the way doesn't
2:22:10
filter right here by the way doesn't this look familiar one one one zero zero
2:22:11
this look familiar one one one zero zero
2:22:11
this look familiar one one one zero zero zero minus one remember that was the
2:22:13
zero minus one remember that was the
2:22:13
zero minus one remember that was the same thing we invented before but here's
2:22:14
same thing we invented before but here's
2:22:14
same thing we invented before but here's a filter
2:22:15
a filter
2:22:16
a filter or a convolution over an important
2:22:17
or a convolution over an important
2:22:17
or a convolution over an important picture obviously the wedding picture
2:22:19
picture obviously the wedding picture
2:22:19
picture obviously the wedding picture super important notice what a
2:22:21
super important notice what a
2:22:21
super important notice what a convolution does to this picture
2:22:24
convolution does to this picture
2:22:24
convolution does to this picture pow notice that this convolution
2:22:26
pow notice that this convolution
2:22:26
pow notice that this convolution actually creates edges
2:22:28
actually creates edges
2:22:28
actually creates edges and then there's this other thing called
2:22:29
and then there's this other thing called
2:22:29
and then there's this other thing called pooling which makes it enhance it right
2:22:32
pooling which makes it enhance it right
2:22:32
pooling which makes it enhance it right so with these numbers i was able to go
2:22:35
so with these numbers i was able to go
2:22:35
so with these numbers i was able to go over
2:22:35
over
2:22:36
over the picture and create edges around it
2:22:38
the picture and create edges around it
2:22:38
the picture and create edges around it let's try let's try one like
2:22:40
let's try let's try one like
2:22:40
let's try let's try one like one of the kids because obviously the
2:22:41
one of the kids because obviously the
2:22:41
one of the kids because obviously the kids are super important here's the kids
2:22:43
kids are super important here's the kids
2:22:43
kids are super important here's the kids right notice that we are going to do a
2:22:46
right notice that we are going to do a
2:22:46
right notice that we are going to do a an image notice
2:22:47
an image notice
2:22:47
an image notice there are the edges and there are the
2:22:50
there are the edges and there are the
2:22:50
there are the edges and there are the enhancement now what if the computer
2:22:52
enhancement now what if the computer
2:22:52
enhancement now what if the computer could come up with its own convolutions
2:22:54
could come up with its own convolutions
2:22:54
could come up with its own convolutions to do crazy things like for example
2:22:56
to do crazy things like for example
2:22:56
to do crazy things like for example here's a here's a serious image so
2:22:58
here's a here's a serious image so
2:22:58
here's a here's a serious image so here's the fence
2:23:02
notice that with the fence it's able to
2:23:04
notice that with the fence it's able to
2:23:04
notice that with the fence it's able to find
2:23:05
find
2:23:05
find exactly where the things are going on
2:23:07
exactly where the things are going on
2:23:07
exactly where the things are going on and i just
2:23:08
and i just
2:23:08
and i just chose these numbers right myself just
2:23:11
chose these numbers right myself just
2:23:11
chose these numbers right myself just like we chose
2:23:12
like we chose
2:23:12
like we chose the numbers before so the question is
2:23:14
the numbers before so the question is
2:23:14
the numbers before so the question is well how do we get
2:23:15
well how do we get
2:23:15
well how do we get all of these w's and b's without us
2:23:18
all of these w's and b's without us
2:23:18
all of these w's and b's without us having to choose them indiscriminately
2:23:19
having to choose them indiscriminately
2:23:19
having to choose them indiscriminately because these images
2:23:20
because these images
2:23:20
because these images can be super rare now i'm using i'm
2:23:22
can be super rare now i'm using i'm
2:23:22
can be super rare now i'm using i'm using the computer vision
2:23:24
using the computer vision
2:23:24
using the computer vision example but this also works with with
2:23:26
example but this also works with with
2:23:26
example but this also works with with all the other things because eventually
2:23:27
all the other things because eventually
2:23:27
all the other things because eventually you have to convert these all the
2:23:28
you have to convert these all the
2:23:28
you have to convert these all the vectors and matrices
2:23:29
vectors and matrices
2:23:29
vectors and matrices well we want to minimize the mistakes
2:23:31
well we want to minimize the mistakes
2:23:31
well we want to minimize the mistakes and so we create this loss function
2:23:33
and so we create this loss function
2:23:33
and so we create this loss function this is called the mean squared error
2:23:35
this is called the mean squared error
2:23:35
this is called the mean squared error and it looks complicated but it's not
2:23:37
and it looks complicated but it's not
2:23:37
and it looks complicated but it's not mean squared error basically means if i
2:23:39
mean squared error basically means if i
2:23:39
mean squared error basically means if i were to take the answer
2:23:41
were to take the answer
2:23:41
were to take the answer right let's just say we guessed the
2:23:43
right let's just say we guessed the
2:23:43
right let's just say we guessed the number three
2:23:44
number three
2:23:44
number three and the real answer is actually 3.
2:23:47
and the real answer is actually 3.
2:23:47
and the real answer is actually 3. notice that
2:23:48
notice that
2:23:48
notice that this here equals 0. but anytime we guess
2:23:51
this here equals 0. but anytime we guess
2:23:51
this here equals 0. but anytime we guess something
2:23:52
something
2:23:52
something that's not right we get something that's
2:23:54
that's not right we get something that's
2:23:54
that's not right we get something that's not 0
2:23:55
not 0
2:23:55
not 0 and then we square it and this loss
2:23:57
and then we square it and this loss
2:23:57
and then we square it and this loss function basically says how
2:23:59
function basically says how
2:23:59
function basically says how bad are we at being predicting the right
2:24:02
bad are we at being predicting the right
2:24:02
bad are we at being predicting the right thing
2:24:02
thing
2:24:02
thing right so it's called the sucky function
2:24:04
right so it's called the sucky function
2:24:04
right so it's called the sucky function in non-specific terms
2:24:06
in non-specific terms
2:24:06
in non-specific terms okay notice that again this is the model
2:24:08
okay notice that again this is the model
2:24:08
okay notice that again this is the model that we have w transpose x plus b
2:24:10
that we have w transpose x plus b
2:24:10
that we have w transpose x plus b minus y and now what we can do is we can
2:24:13
minus y and now what we can do is we can
2:24:13
minus y and now what we can do is we can use
2:24:14
use
2:24:14
use some special maths to figure out what
2:24:17
some special maths to figure out what
2:24:17
some special maths to figure out what exactly the appropriate w's and b's are
2:24:20
exactly the appropriate w's and b's are
2:24:20
exactly the appropriate w's and b's are and we have to resort to calculus now
2:24:21
and we have to resort to calculus now
2:24:22
and we have to resort to calculus now here's the thing this
2:24:23
here's the thing this
2:24:23
here's the thing this little doohickey here has a square here
2:24:25
little doohickey here has a square here
2:24:25
little doohickey here has a square here so we're going to pretend that it's
2:24:26
so we're going to pretend that it's
2:24:26
so we're going to pretend that it's something like this
2:24:27
something like this
2:24:27
something like this and notice that in in school you
2:24:29
and notice that in in school you
2:24:29
and notice that in in school you remember this particular function
2:24:31
remember this particular function
2:24:31
remember this particular function called the parabola now imagine you're
2:24:33
called the parabola now imagine you're
2:24:33
called the parabola now imagine you're like a blind mario
2:24:35
like a blind mario
2:24:35
like a blind mario you can't see right mario lives in 2d
2:24:37
you can't see right mario lives in 2d
2:24:37
you can't see right mario lives in 2d space and you want to know
2:24:38
space and you want to know
2:24:38
space and you want to know you want to know where to walk to get
2:24:41
you want to know where to walk to get
2:24:41
you want to know where to walk to get the optimal value which is zero right
2:24:43
the optimal value which is zero right
2:24:43
the optimal value which is zero right because we want this all to equal to
2:24:44
because we want this all to equal to
2:24:44
because we want this all to equal to zero we want to set up
2:24:46
zero we want to set up
2:24:46
zero we want to set up set it up in such a way so that this
2:24:48
set it up in such a way so that this
2:24:48
set it up in such a way so that this equals zero well how do you do that well
2:24:50
equals zero well how do you do that well
2:24:50
equals zero well how do you do that well mario needs to walk down this way
2:24:54
mario needs to walk down this way
2:24:54
mario needs to walk down this way until he gets to here right so how does
2:24:57
until he gets to here right so how does
2:24:57
until he gets to here right so how does he know well
2:24:58
he know well
2:24:58
he know well why don't we do this um why don't we do
2:25:01
why don't we do this um why don't we do
2:25:01
why don't we do this um why don't we do this why don't we just tell him like
2:25:03
this why don't we just tell him like
2:25:03
this why don't we just tell him like like if you're if you're mario what you
2:25:05
like if you're if you're mario what you
2:25:05
like if you're if you're mario what you would do is you'd put your foot a little
2:25:06
would do is you'd put your foot a little
2:25:06
would do is you'd put your foot a little bit
2:25:06
bit
2:25:06
bit forward and a little bit backwards and
2:25:08
forward and a little bit backwards and
2:25:08
forward and a little bit backwards and then what you would do is you would
2:25:10
then what you would do is you would
2:25:10
then what you would do is you would measure
2:25:11
measure
2:25:11
measure like which place is higher right so
2:25:13
like which place is higher right so
2:25:13
like which place is higher right so we're taking a little bit of the x let's
2:25:15
we're taking a little bit of the x let's
2:25:15
we're taking a little bit of the x let's just say dx
2:25:16
just say dx
2:25:16
just say dx and we're taking a little bit of the y d
2:25:18
and we're taking a little bit of the y d
2:25:18
and we're taking a little bit of the y d y and what we're going to do is we're
2:25:19
y and what we're going to do is we're
2:25:19
y and what we're going to do is we're going to divide them d y by d
2:25:21
going to divide them d y by d
2:25:21
going to divide them d y by d x right and then what we're going to do
2:25:23
x right and then what we're going to do
2:25:23
x right and then what we're going to do is we're going to take the limit as this
2:25:25
is we're going to take the limit as this
2:25:25
is we're going to take the limit as this distance
2:25:25
distance
2:25:25
distance h goes to 0 and this effectively becomes
2:25:28
h goes to 0 and this effectively becomes
2:25:28
h goes to 0 and this effectively becomes what's called the derivative and the
2:25:30
what's called the derivative and the
2:25:30
what's called the derivative and the derivative tells you
2:25:31
derivative tells you
2:25:31
derivative tells you the slope which we should go into
2:25:34
the slope which we should go into
2:25:34
the slope which we should go into to do that so how do we do that right
2:25:37
to do that so how do we do that right
2:25:37
to do that so how do we do that right well it turns out that pie torch tensors
2:25:39
well it turns out that pie torch tensors
2:25:39
well it turns out that pie torch tensors are
2:25:39
are
2:25:39
are actually super special let me show you
2:25:42
actually super special let me show you
2:25:42
actually super special let me show you why
2:25:43
why
2:25:43
why uh let me go back to my this thing right
2:25:45
uh let me go back to my this thing right
2:25:45
uh let me go back to my this thing right here
2:25:46
here
2:25:46
here so pi torch tensors are actually really
2:25:49
so pi torch tensors are actually really
2:25:49
so pi torch tensors are actually really interesting because
2:25:50
interesting because
2:25:50
interesting because when you when you are creating so here's
2:25:52
when you when you are creating so here's
2:25:52
when you when you are creating so here's the here's the
2:25:53
the here's the
2:25:53
the here's the tensors right notice here's our w and
2:25:55
tensors right notice here's our w and
2:25:55
tensors right notice here's our w and here's our b
2:25:56
here's our b
2:25:56
here's our b we're taking x matrix multiplying it by
2:25:59
we're taking x matrix multiplying it by
2:25:59
we're taking x matrix multiplying it by w adding the b
2:26:00
w adding the b
2:26:00
w adding the b here is the loss function notice what pi
2:26:02
here is the loss function notice what pi
2:26:02
here is the loss function notice what pi torch does is with any function
2:26:04
torch does is with any function
2:26:04
torch does is with any function it actually keeps track of the variables
2:26:07
it actually keeps track of the variables
2:26:07
it actually keeps track of the variables that you want to
2:26:08
that you want to
2:26:08
that you want to retain the derivatives for derivatives
2:26:10
retain the derivatives for derivatives
2:26:10
retain the derivatives for derivatives in multiple directions are called
2:26:12
in multiple directions are called
2:26:12
in multiple directions are called gradients
2:26:13
gradients
2:26:13
gradients and so that's what it does and it turns
2:26:15
and so that's what it does and it turns
2:26:15
and so that's what it does and it turns out that
2:26:16
out that
2:26:16
out that in its internal state it actually
2:26:18
in its internal state it actually
2:26:18
in its internal state it actually remembers what the derivatives are
2:26:20
remembers what the derivatives are
2:26:20
remembers what the derivatives are and it knows how to iteratively as you
2:26:23
and it knows how to iteratively as you
2:26:23
and it knows how to iteratively as you go in a loop
2:26:24
go in a loop
2:26:24
go in a loop it calculates the gradients and then it
2:26:25
it calculates the gradients and then it
2:26:25
it calculates the gradients and then it knows how to walk
2:26:27
knows how to walk
2:26:27
knows how to walk in the right direction uh and for
2:26:29
in the right direction uh and for
2:26:29
in the right direction uh and for walking in the right direction basically
2:26:31
walking in the right direction basically
2:26:31
walking in the right direction basically what it does
2:26:31
what it does
2:26:31
what it does if you take the gradients and you
2:26:33
if you take the gradients and you
2:26:33
if you take the gradients and you subtract them off the w and then you
2:26:34
subtract them off the w and then you
2:26:34
subtract them off the w and then you keep doing that until you reach
2:26:36
keep doing that until you reach
2:26:36
keep doing that until you reach some optimal state and notice that all
2:26:38
some optimal state and notice that all
2:26:38
some optimal state and notice that all you need to do to figure out the
2:26:39
you need to do to figure out the
2:26:39
you need to do to figure out the gradients is through loss
2:26:41
gradients is through loss
2:26:41
gradients is through loss dot backward and all you need to do is
2:26:42
dot backward and all you need to do is
2:26:42
dot backward and all you need to do is subtract this
2:26:44
subtract this
2:26:44
subtract this from the current w's and b's and you
2:26:45
from the current w's and b's and you
2:26:45
from the current w's and b's and you keep doing this until their loss goes to
2:26:48
keep doing this until their loss goes to
2:26:48
keep doing this until their loss goes to as close to zero as you can when you're
2:26:50
as close to zero as you can when you're
2:26:50
as close to zero as you can when you're setting these up in pytorch you can
2:26:51
setting these up in pytorch you can
2:26:51
setting these up in pytorch you can actually create
2:26:52
actually create
2:26:52
actually create really cool models like here's a linear
2:26:54
really cool models like here's a linear
2:26:54
really cool models like here's a linear model
2:26:55
model
2:26:55
model right uh here is a neural network model
2:26:59
right uh here is a neural network model
2:26:59
right uh here is a neural network model right it keeps all of this and because
2:27:01
right it keeps all of this and because
2:27:01
right it keeps all of this and because add and matrix multiply happens so much
2:27:03
add and matrix multiply happens so much
2:27:03
add and matrix multiply happens so much there's basically a single op to do that
2:27:05
there's basically a single op to do that
2:27:05
there's basically a single op to do that which is really cool
2:27:05
which is really cool
2:27:06
which is really cool and here is this convolutional neural
2:27:08
and here is this convolutional neural
2:27:08
and here is this convolutional neural network
2:27:09
network
2:27:09
network right you can see it it starts to get
2:27:11
right you can see it it starts to get
2:27:11
right you can see it it starts to get deeper and deeper until you get to
2:27:13
deeper and deeper until you get to
2:27:13
deeper and deeper until you get to the actual model that we're gonna use
2:27:17
the actual model that we're gonna use
2:27:17
the actual model that we're gonna use which is called
2:27:18
which is called
2:27:18
which is called mobilenet right notice how deep this one
2:27:20
mobilenet right notice how deep this one
2:27:20
mobilenet right notice how deep this one is and this is the actual model
2:27:23
is and this is the actual model
2:27:23
is and this is the actual model that we learned to do tacos versus
2:27:25
that we learned to do tacos versus
2:27:25
that we learned to do tacos versus burritos okay
2:27:26
burritos okay
2:27:26
burritos okay so if i were to go in here uh let's go
2:27:29
so if i were to go in here uh let's go
2:27:29
so if i were to go in here uh let's go to uh some numbers this is a model
2:27:30
to uh some numbers this is a model
2:27:30
to uh some numbers this is a model that's already trained
2:27:31
that's already trained
2:27:31
that's already trained notice you have this gmm it stands for
2:27:34
notice you have this gmm it stands for
2:27:34
notice you have this gmm it stands for generic
2:27:35
generic
2:27:35
generic mult matrix multiplication plus the
2:27:37
mult matrix multiplication plus the
2:27:37
mult matrix multiplication plus the addition of the bias
2:27:38
addition of the bias
2:27:38
addition of the bias here you have the actual matrix right
2:27:42
here you have the actual matrix right
2:27:42
here you have the actual matrix right two by a thousand and it turns out
2:27:44
two by a thousand and it turns out
2:27:44
two by a thousand and it turns out remember how i told you that that the
2:27:45
remember how i told you that that the
2:27:45
remember how i told you that that the neural networks get narrower as you go
2:27:47
neural networks get narrower as you go
2:27:47
neural networks get narrower as you go down
2:27:47
down
2:27:47
down notice that the output vector is gonna
2:27:50
notice that the output vector is gonna
2:27:50
notice that the output vector is gonna be of size two
2:27:51
be of size two
2:27:51
be of size two because we're only guessing between
2:27:53
because we're only guessing between
2:27:53
because we're only guessing between tacos and burritos by math this is how
2:27:56
tacos and burritos by math this is how
2:27:56
tacos and burritos by math this is how we're doing it
2:27:56
we're doing it
2:27:56
we're doing it and notice that there's a thousand
2:27:58
and notice that there's a thousand
2:27:58
and notice that there's a thousand inputs coming and then it's gonna be
2:28:00
inputs coming and then it's gonna be
2:28:00
inputs coming and then it's gonna be whittled down to two and then we have
2:28:02
whittled down to two and then we have
2:28:02
whittled down to two and then we have this soft max function the softmax
2:28:04
this soft max function the softmax
2:28:04
this soft max function the softmax function
2:28:05
function
2:28:05
function forces it to sum to one so it looks like
2:28:07
forces it to sum to one so it looks like
2:28:07
forces it to sum to one so it looks like a percentage
2:28:09
a percentage
2:28:09
a percentage okay so that's what's going on there and
2:28:11
okay so that's what's going on there and
2:28:11
okay so that's what's going on there and now when we go back to it
2:28:13
now when we go back to it
2:28:13
now when we go back to it if you're looking at how it predicts the
2:28:16
if you're looking at how it predicts the
2:28:16
if you're looking at how it predicts the pizza
2:28:17
pizza
2:28:17
pizza notice that the majority of the weight
2:28:20
notice that the majority of the weight
2:28:20
notice that the majority of the weight landed in burrito
2:28:22
landed in burrito
2:28:22
landed in burrito because that's all it could predict it
2:28:24
because that's all it could predict it
2:28:24
because that's all it could predict it had
2:28:25
had
2:28:25
had no other notion of anything else because
2:28:27
no other notion of anything else because
2:28:27
no other notion of anything else because when i train the model
2:28:29
when i train the model
2:28:29
when i train the model which i do right here notice that i only
2:28:31
which i do right here notice that i only
2:28:31
which i do right here notice that i only give it
2:28:32
give it
2:28:32
give it burritos and tacos right sorry right
2:28:35
burritos and tacos right sorry right
2:28:35
burritos and tacos right sorry right here
2:28:36
here
2:28:36
here burritos and tacos those are the classes
2:28:39
burritos and tacos those are the classes
2:28:39
burritos and tacos those are the classes and you can see that i load up the model
2:28:40
and you can see that i load up the model
2:28:40
and you can see that i load up the model here
2:28:41
here
2:28:41
here uh here's the sequential sequential
2:28:43
uh here's the sequential sequential
2:28:43
uh here's the sequential sequential means do these things in order
2:28:44
means do these things in order
2:28:44
means do these things in order i just stole a mobilenet model added the
2:28:47
i just stole a mobilenet model added the
2:28:47
i just stole a mobilenet model added the activation function i talked about
2:28:49
activation function i talked about
2:28:49
activation function i talked about here is the another w transpose x times
2:28:51
here is the another w transpose x times
2:28:51
here is the another w transpose x times b which gets this and then we do the
2:28:53
b which gets this and then we do the
2:28:53
b which gets this and then we do the soft max
2:28:54
soft max
2:28:54
soft max here is the loss function we're using
2:28:56
here is the loss function we're using
2:28:56
here is the loss function we're using binary i should i probably should use
2:28:57
binary i should i probably should use
2:28:57
binary i should i probably should use binary cross entropy
2:28:59
binary cross entropy
2:28:59
binary cross entropy it probably would have been better but
2:29:00
it probably would have been better but
2:29:00
it probably would have been better but here i'm using cross entropy and you're
2:29:02
here i'm using cross entropy and you're
2:29:02
here i'm using cross entropy and you're probably wondering well how did you went
2:29:03
probably wondering well how did you went
2:29:03
probably wondering well how did you went from
2:29:04
from
2:29:04
from uh you went from mean squared error
2:29:06
uh you went from mean squared error
2:29:06
uh you went from mean squared error which was the the you know the h of x
2:29:08
which was the the you know the h of x
2:29:08
which was the the you know the h of x minus y squared
2:29:09
minus y squared
2:29:09
minus y squared there are other loss functions that
2:29:10
there are other loss functions that
2:29:10
there are other loss functions that measure suckiness and you need to use
2:29:12
measure suckiness and you need to use
2:29:12
measure suckiness and you need to use the right one for the right problem
2:29:14
the right one for the right problem
2:29:14
the right one for the right problem and then here is sgd which stands for
2:29:16
and then here is sgd which stands for
2:29:16
and then here is sgd which stands for stochastic gradient descent which is
2:29:18
stochastic gradient descent which is
2:29:18
stochastic gradient descent which is basically the mario
2:29:19
basically the mario
2:29:19
basically the mario like jumping around to the bottom of the
2:29:22
like jumping around to the bottom of the
2:29:22
like jumping around to the bottom of the pit right
2:29:23
pit right
2:29:23
pit right and here's the train model right and
2:29:26
and here's the train model right and
2:29:26
and here's the train model right and when i go to it let me go to the
2:29:27
when i go to it let me go to the
2:29:27
when i go to it let me go to the definition here you can see basically
2:29:29
definition here you can see basically
2:29:29
definition here you can see basically i'm just doing a loop where i optimize
2:29:33
i'm just doing a loop where i optimize
2:29:33
i'm just doing a loop where i optimize everything
2:29:33
everything
2:29:33
everything i zero the gradients right if we're
2:29:35
i zero the gradients right if we're
2:29:35
i zero the gradients right if we're training right i
2:29:37
training right i
2:29:37
training right i run the model function i figure out the
2:29:40
run the model function i figure out the
2:29:40
run the model function i figure out the loss
2:29:41
loss
2:29:41
loss and then i i go i do the backward step
2:29:44
and then i i go i do the backward step
2:29:44
and then i i go i do the backward step which finds the gradient and then i step
2:29:46
which finds the gradient and then i step
2:29:46
which finds the gradient and then i step and i do this over and over and over and
2:29:47
and i do this over and over and over and
2:29:47
and i do this over and over and over and over until it solves
2:29:49
over until it solves
2:29:49
over until it solves the actual problem and then you get
2:29:53
the actual problem and then you get
2:29:53
the actual problem and then you get um this
2:29:56
um this
2:29:56
um this okay so i went super fast through like a
2:29:59
okay so i went super fast through like a
2:29:59
okay so i went super fast through like a lot of maths
2:30:00
lot of maths
2:30:00
lot of maths and a lot of neural network stuff but
2:30:03
and a lot of neural network stuff but
2:30:03
and a lot of neural network stuff but the fundamental
2:30:03
the fundamental
2:30:04
the fundamental thing to understand is that there are
2:30:05
thing to understand is that there are
2:30:06
thing to understand is that there are three things that are happening
2:30:07
three things that are happening
2:30:07
three things that are happening you have to construct a model function
2:30:10
you have to construct a model function
2:30:10
you have to construct a model function notice that we decided to use mobilenet
2:30:12
notice that we decided to use mobilenet
2:30:12
notice that we decided to use mobilenet which looks like this
2:30:14
which looks like this
2:30:14
which looks like this right we started earlier with just a
2:30:15
right we started earlier with just a
2:30:15
right we started earlier with just a general like w transpose x plus b
2:30:18
general like w transpose x plus b
2:30:18
general like w transpose x plus b notice that we can start to stack them
2:30:20
notice that we can start to stack them
2:30:20
notice that we can start to stack them with functions in between
2:30:21
with functions in between
2:30:22
with functions in between and then you can start to do more and
2:30:23
and then you can start to do more and
2:30:23
and then you can start to do more and then these things get bigger and bigger
2:30:24
then these things get bigger and bigger
2:30:24
then these things get bigger and bigger and bigger
2:30:25
and bigger
2:30:26
and bigger but i want to be completely transparent
2:30:28
but i want to be completely transparent
2:30:28
but i want to be completely transparent and tell you that these things
2:30:30
and tell you that these things
2:30:30
and tell you that these things are not self-aware they have no idea
2:30:32
are not self-aware they have no idea
2:30:32
are not self-aware they have no idea what's going on
2:30:33
what's going on
2:30:33
what's going on they are basically it's like a rock
2:30:37
they are basically it's like a rock
2:30:37
they are basically it's like a rock right that humans drew smiley faces on
2:30:39
right that humans drew smiley faces on
2:30:39
right that humans drew smiley faces on that make it look like it thinks it is
2:30:41
that make it look like it thinks it is
2:30:41
that make it look like it thinks it is because right now we think wow it's just
2:30:43
because right now we think wow it's just
2:30:43
because right now we think wow it's just it has like a really funny joke it
2:30:45
it has like a really funny joke it
2:30:45
it has like a really funny joke it thinks that a pizza is a burrito
2:30:47
thinks that a pizza is a burrito
2:30:47
thinks that a pizza is a burrito no it doesn't think anything all it did
2:30:48
no it doesn't think anything all it did
2:30:48
no it doesn't think anything all it did is it multiplied all these pixels
2:30:51
is it multiplied all these pixels
2:30:51
is it multiplied all these pixels by certain numbers ran functions over
2:30:53
by certain numbers ran functions over
2:30:53
by certain numbers ran functions over them and did it deeply
2:30:54
them and did it deeply
2:30:54
them and did it deeply until it got to an answer at the bottom
2:30:56
until it got to an answer at the bottom
2:30:56
until it got to an answer at the bottom and so that's why i don't
2:30:57
and so that's why i don't
2:30:57
and so that's why i don't i don't like when we anthropomorphize ai
2:31:00
i don't like when we anthropomorphize ai
2:31:00
i don't like when we anthropomorphize ai because what it does is it shifts the
2:31:02
because what it does is it shifts the
2:31:02
because what it does is it shifts the responsibility
2:31:04
responsibility
2:31:04
responsibility of ethics from the humans to the model
2:31:07
of ethics from the humans to the model
2:31:07
of ethics from the humans to the model the model has no sentient
2:31:10
the model has no sentient
2:31:10
the model has no sentient ability at all when it comes to agi
2:31:12
ability at all when it comes to agi
2:31:12
ability at all when it comes to agi which is which is
2:31:13
which is which is
2:31:14
which is which is um you know generalized artificial
2:31:15
um you know generalized artificial
2:31:16
um you know generalized artificial intelligence we are nowhere near
2:31:18
intelligence we are nowhere near
2:31:18
intelligence we are nowhere near anywhere like that and that's where i
2:31:21
anywhere like that and that's where i
2:31:21
anywhere like that and that's where i think
2:31:21
think
2:31:21
think a lot of the ethical issues come on
2:31:23
a lot of the ethical issues come on
2:31:23
a lot of the ethical issues come on because if someone ever tells you well
2:31:24
because if someone ever tells you well
2:31:24
because if someone ever tells you well that's what the model predicted
2:31:25
that's what the model predicted
2:31:26
that's what the model predicted you should say hold on wait a minute
2:31:29
you should say hold on wait a minute
2:31:29
you should say hold on wait a minute are you giving the model a pizza when it
2:31:32
are you giving the model a pizza when it
2:31:32
are you giving the model a pizza when it can only detect tacos and burritos
2:31:34
can only detect tacos and burritos
2:31:34
can only detect tacos and burritos one question another question what kind
2:31:36
one question another question what kind
2:31:36
one question another question what kind of pictures did you use did you use
2:31:39
of pictures did you use did you use
2:31:39
of pictures did you use did you use actual burritos or did you use pictures
2:31:41
actual burritos or did you use pictures
2:31:41
actual burritos or did you use pictures of
2:31:42
of
2:31:42
of literal burritos which i in fact
2:31:45
literal burritos which i in fact
2:31:45
literal burritos which i in fact did when i trained it let me go
2:31:48
did when i trained it let me go
2:31:48
did when i trained it let me go up to this like i literally used
2:31:51
up to this like i literally used
2:31:51
up to this like i literally used a real burrito when i trained it right
2:31:54
a real burrito when i trained it right
2:31:54
a real burrito when i trained it right that's funny
2:31:55
that's funny
2:31:55
that's funny but still and so that's where the
2:31:58
but still and so that's where the
2:31:58
but still and so that's where the ethical
2:31:58
ethical
2:31:58
ethical questions i think lie for me when it
2:32:01
questions i think lie for me when it
2:32:01
questions i think lie for me when it comes to building for example computer
2:32:03
comes to building for example computer
2:32:03
comes to building for example computer vision models you are
2:32:04
vision models you are
2:32:04
vision models you are effectively using multi-variable
2:32:07
effectively using multi-variable
2:32:07
effectively using multi-variable differential calculus
2:32:08
differential calculus
2:32:08
differential calculus to iteratively create the best
2:32:12
to iteratively create the best
2:32:12
to iteratively create the best possible numbers weights and and biases
2:32:15
possible numbers weights and and biases
2:32:15
possible numbers weights and and biases in order to optimize the loss
2:32:18
in order to optimize the loss
2:32:18
in order to optimize the loss function that's it and you're probably
2:32:21
function that's it and you're probably
2:32:21
function that's it and you're probably what about the huge models that actually
2:32:23
what about the huge models that actually
2:32:23
what about the huge models that actually generate stuff it's the same thing
2:32:26
generate stuff it's the same thing
2:32:26
generate stuff it's the same thing except we're using for example for
2:32:28
except we're using for example for
2:32:28
except we're using for example for for for uh generative uh generative
2:32:30
for for uh generative uh generative
2:32:30
for for uh generative uh generative models like general adversarial networks
2:32:32
models like general adversarial networks
2:32:32
models like general adversarial networks uh they're basically two networks that
2:32:34
uh they're basically two networks that
2:32:34
uh they're basically two networks that are fighting against each other to
2:32:35
are fighting against each other to
2:32:36
are fighting against each other to generate
2:32:36
generate
2:32:36
generate more stuff or for example for generative
2:32:38
more stuff or for example for generative
2:32:38
more stuff or for example for generative models uh the transpose of the
2:32:40
models uh the transpose of the
2:32:40
models uh the transpose of the convolution will generate images
2:32:42
convolution will generate images
2:32:42
convolution will generate images for you but it's still using
2:32:44
for you but it's still using
2:32:44
for you but it's still using optimization to figure out
2:32:46
optimization to figure out
2:32:46
optimization to figure out the loss between the actual pictures and
2:32:48
the loss between the actual pictures and
2:32:48
the loss between the actual pictures and the ones that are generated
2:32:50
the ones that are generated
2:32:50
the ones that are generated it's still all just math there's no
2:32:53
it's still all just math there's no
2:32:53
it's still all just math there's no magicalness at all and by the way this
2:32:55
magicalness at all and by the way this
2:32:55
magicalness at all and by the way this code is available you can
2:32:57
code is available you can
2:32:57
code is available you can you can go look at it let me bring it
2:32:58
you can go look at it let me bring it
2:32:58
you can go look at it let me bring it over here so you can
2:33:00
over here so you can
2:33:00
over here so you can find it so github.com
2:33:03
find it so github.com
2:33:03
find it so github.com uh i think it's uh the the one that i
2:33:06
uh i think it's uh the the one that i
2:33:06
uh i think it's uh the the one that i was showing you that has the
2:33:07
was showing you that has the
2:33:07
was showing you that has the explanations
2:33:08
explanations
2:33:08
explanations is deep learning with pytorch so let me
2:33:10
is deep learning with pytorch so let me
2:33:10
is deep learning with pytorch so let me bring that over here so you can see that
2:33:13
bring that over here so you can see that
2:33:13
bring that over here so you can see that so so those kernels right that i was
2:33:15
so so those kernels right that i was
2:33:15
so so those kernels right that i was running and those those things that show
2:33:16
running and those those things that show
2:33:16
running and those those things that show you
2:33:17
you
2:33:17
you like the actual gr execution graph those
2:33:20
like the actual gr execution graph those
2:33:20
like the actual gr execution graph those are all there
2:33:20
are all there
2:33:20
are all there and then you can look at the tacos
2:33:22
and then you can look at the tacos
2:33:22
and then you can look at the tacos versus burritos if you go to food ai
2:33:25
versus burritos if you go to food ai
2:33:25
versus burritos if you go to food ai right you're going to find all of that
2:33:26
right you're going to find all of that
2:33:26
right you're going to find all of that code right there so that code that i
2:33:27
code right there so that code that i
2:33:27
code right there so that code that i just showed you
2:33:28
just showed you
2:33:28
just showed you is right here right i am making some
2:33:31
is right here right i am making some
2:33:31
is right here right i am making some changes so you can see on the dev branch
2:33:33
changes so you can see on the dev branch
2:33:33
changes so you can see on the dev branch i'm trying to organize things a little
2:33:35
i'm trying to organize things a little
2:33:35
i'm trying to organize things a little bit better and so that's what you
2:33:37
bit better and so that's what you
2:33:37
bit better and so that's what you actually saw my code doing well i
2:33:38
actually saw my code doing well i
2:33:38
actually saw my code doing well i haven't checked it in yet
2:33:40
haven't checked it in yet
2:33:40
haven't checked it in yet oh i need to go to dev sorry so you can
2:33:42
oh i need to go to dev sorry so you can
2:33:42
oh i need to go to dev sorry so you can see that i'm starting to
2:33:43
see that i'm starting to
2:33:43
see that i'm starting to make it a little bit more sensible right
2:33:46
make it a little bit more sensible right
2:33:46
make it a little bit more sensible right and then
2:33:46
and then
2:33:46
and then add more explanations but it's all there
2:33:49
add more explanations but it's all there
2:33:49
add more explanations but it's all there uh there
2:33:49
uh there
2:33:49
uh there is the food ai and then deep learning
2:33:52
is the food ai and then deep learning
2:33:52
is the food ai and then deep learning with pie torch
2:33:53
with pie torch
2:33:53
with pie torch okay let's go to the questions because
2:33:57
okay let's go to the questions because
2:33:57
okay let's go to the questions because i went super fast let me turn here
2:34:01
i went super fast let me turn here
2:34:01
i went super fast let me turn here okay how easy is okay no i need to
2:34:04
okay how easy is okay no i need to
2:34:04
okay how easy is okay no i need to scroll down
2:34:05
scroll down
2:34:05
scroll down okay any questions so far
2:34:09
okay any questions so far
2:34:09
okay any questions so far wow you should definitely take a
2:34:11
wow you should definitely take a
2:34:11
wow you should definitely take a breather that's that's like
2:34:12
breather that's that's like
2:34:12
breather that's that's like four years of mass in in 30 minutes
2:34:15
four years of mass in in 30 minutes
2:34:15
four years of mass in in 30 minutes that's amazing
2:34:18
i definitely was having flashbacks to
2:34:20
i definitely was having flashbacks to
2:34:20
i definitely was having flashbacks to college so thank you seth
2:34:24
that was a lot of god during the three
2:34:26
that was a lot of god during the three
2:34:26
that was a lot of god during the three o'clock hour for me
2:34:30
o'clock hour for me
2:34:30
o'clock hour for me man yeah a huge flashback
2:34:33
man yeah a huge flashback
2:34:33
man yeah a huge flashback everyone's like like sitting on the
2:34:35
everyone's like like sitting on the
2:34:35
everyone's like like sitting on the floor like by the way if you see me
2:34:37
floor like by the way if you see me
2:34:37
floor like by the way if you see me turning my head i'm looking in the chat
2:34:39
turning my head i'm looking in the chat
2:34:40
turning my head i'm looking in the chat nope i should holy cow maybe i'm such a
2:34:42
nope i should holy cow maybe i'm such a
2:34:42
nope i should holy cow maybe i'm such a good
2:34:48
so it was kind of interesting how you
2:34:50
so it was kind of interesting how you
2:34:50
so it was kind of interesting how you kind of simplified it down to
2:34:51
kind of simplified it down to
2:34:52
kind of simplified it down to you know are you passing a picture of a
2:34:55
you know are you passing a picture of a
2:34:55
you know are you passing a picture of a pizza
2:34:55
pizza
2:34:55
pizza into your model when you've only trained
2:34:58
into your model when you've only trained
2:34:58
into your model when you've only trained it to recognize burritos and tacos
2:35:00
it to recognize burritos and tacos
2:35:00
it to recognize burritos and tacos so how common is that when people build
2:35:02
so how common is that when people build
2:35:02
so how common is that when people build models because i
2:35:03
models because i
2:35:03
models because i it sounds like a very common sense
2:35:06
it sounds like a very common sense
2:35:06
it sounds like a very common sense question but it truly sounds like
2:35:08
question but it truly sounds like
2:35:08
question but it truly sounds like something that we're doing because
2:35:09
something that we're doing because
2:35:09
something that we're doing because when we look at our data sets our data
2:35:12
when we look at our data sets our data
2:35:12
when we look at our data sets our data sets
2:35:13
sets
2:35:13
sets are comprised of what we think to be
2:35:17
are comprised of what we think to be
2:35:17
are comprised of what we think to be the true positive but how good are we at
2:35:20
the true positive but how good are we at
2:35:20
the true positive but how good are we at actually
2:35:21
actually
2:35:21
actually kind of starting the problem statement
2:35:23
kind of starting the problem statement
2:35:23
kind of starting the problem statement off with the correct set of data
2:35:26
off with the correct set of data
2:35:26
off with the correct set of data yeah i don't know i mean it depends
2:35:28
yeah i don't know i mean it depends
2:35:28
yeah i don't know i mean it depends right especially when you have
2:35:30
right especially when you have
2:35:30
right especially when you have um when you have a ton of data that
2:35:32
um when you have a ton of data that
2:35:32
um when you have a ton of data that that's not being looked at
2:35:34
that's not being looked at
2:35:34
that's not being looked at that's a problem i'll give you an
2:35:35
that's a problem i'll give you an
2:35:35
that's a problem i'll give you an example and this is this this is a silly
2:35:37
example and this is this this is a silly
2:35:37
example and this is this this is a silly one but there's a book called weapons of
2:35:39
one but there's a book called weapons of
2:35:39
one but there's a book called weapons of math
2:35:39
math
2:35:40
math destruction m-a-t-h-h kind of funny
2:35:42
destruction m-a-t-h-h kind of funny
2:35:42
destruction m-a-t-h-h kind of funny right weapons of mathematics
2:35:43
right weapons of mathematics
2:35:44
right weapons of mathematics where are they where they show
2:35:45
where are they where they show
2:35:45
where are they where they show statistics for example if i were to
2:35:46
statistics for example if i were to
2:35:46
statistics for example if i were to train a model
2:35:47
train a model
2:35:47
train a model to tell law enforcement in the united
2:35:49
to tell law enforcement in the united
2:35:49
to tell law enforcement in the united states where to go police
2:35:51
states where to go police
2:35:51
states where to go police well statistically and historically
2:35:55
well statistically and historically
2:35:55
well statistically and historically black americans have been over policed
2:35:57
black americans have been over policed
2:35:58
black americans have been over policed that's just in certain places that's
2:35:59
that's just in certain places that's
2:35:59
that's just in certain places that's just the way it is
2:36:00
just the way it is
2:36:00
just the way it is so if i were to use that data to produce
2:36:03
so if i were to use that data to produce
2:36:03
so if i were to use that data to produce a neural network model
2:36:04
a neural network model
2:36:04
a neural network model it would basically just learn that it
2:36:07
it would basically just learn that it
2:36:07
it would basically just learn that it has to go
2:36:08
has to go
2:36:08
has to go police black americans again right
2:36:11
police black americans again right
2:36:11
police black americans again right another example an nlp
2:36:13
another example an nlp
2:36:13
another example an nlp when you look at these things for
2:36:14
when you look at these things for
2:36:14
when you look at these things for example like for similarity of words
2:36:17
example like for similarity of words
2:36:17
example like for similarity of words or which two words go together it turns
2:36:20
or which two words go together it turns
2:36:20
or which two words go together it turns out that historically
2:36:22
out that historically
2:36:22
out that historically for whatever sexist reasons the term
2:36:25
for whatever sexist reasons the term
2:36:25
for whatever sexist reasons the term doctor and maleness are related in
2:36:29
doctor and maleness are related in
2:36:29
doctor and maleness are related in almost everything we've written up to
2:36:31
almost everything we've written up to
2:36:31
almost everything we've written up to this point
2:36:32
this point
2:36:32
this point and so you would find that an nlp model
2:36:35
and so you would find that an nlp model
2:36:35
and so you would find that an nlp model might suggest that a doctor is a male
2:36:38
might suggest that a doctor is a male
2:36:38
might suggest that a doctor is a male thing versus a female thing
2:36:40
thing versus a female thing
2:36:40
thing versus a female thing where in reality a doctor is just a
2:36:42
where in reality a doctor is just a
2:36:42
where in reality a doctor is just a doctor right and so you'll see a ton of
2:36:44
doctor right and so you'll see a ton of
2:36:44
doctor right and so you'll see a ton of stuff like that
2:36:45
stuff like that
2:36:45
stuff like that inherently built into the data and so
2:36:48
inherently built into the data and so
2:36:48
inherently built into the data and so the reality of the matter is the first
2:36:50
the reality of the matter is the first
2:36:50
the reality of the matter is the first question you should ask before doing any
2:36:51
question you should ask before doing any
2:36:51
question you should ask before doing any of these things
2:36:52
of these things
2:36:52
of these things is number one who is are these
2:36:56
is number one who is are these
2:36:56
is number one who is are these models affecting who who will these
2:36:58
models affecting who who will these
2:36:58
models affecting who who will these decisions affect
2:36:59
decisions affect
2:36:59
decisions affect and number two how might it affect
2:37:02
and number two how might it affect
2:37:02
and number two how might it affect people
2:37:03
people
2:37:03
people unfairly that's what you should start
2:37:05
unfairly that's what you should start
2:37:05
unfairly that's what you should start with and then you need to go into the
2:37:07
with and then you need to go into the
2:37:07
with and then you need to go into the data and look at it
2:37:08
data and look at it
2:37:08
data and look at it and then you need to go into into these
2:37:10
and then you need to go into into these
2:37:10
and then you need to go into into these model uh black box and white box models
2:37:13
model uh black box and white box models
2:37:13
model uh black box and white box models for
2:37:13
for
2:37:13
for responsible explainability that noel was
2:37:15
responsible explainability that noel was
2:37:15
responsible explainability that noel was talking about to verify if these things
2:37:18
talking about to verify if these things
2:37:18
talking about to verify if these things are happening or not and so that's
2:37:19
are happening or not and so that's
2:37:19
are happening or not and so that's that's your goals but the reality again
2:37:21
that's your goals but the reality again
2:37:21
that's your goals but the reality again hopefully you saw even if you didn't get
2:37:22
hopefully you saw even if you didn't get
2:37:22
hopefully you saw even if you didn't get any all of it because like i said i went
2:37:24
any all of it because like i said i went
2:37:24
any all of it because like i said i went through like
2:37:25
through like
2:37:25
through like a year of grad school for the uh for
2:37:27
a year of grad school for the uh for
2:37:27
a year of grad school for the uh for like 20 minutes
2:37:29
like 20 minutes
2:37:29
like 20 minutes you you recognize that it's not doing
2:37:31
you you recognize that it's not doing
2:37:31
you you recognize that it's not doing anything
2:37:32
anything
2:37:32
anything other than learning numbers to multiply
2:37:34
other than learning numbers to multiply
2:37:34
other than learning numbers to multiply the input by
2:37:35
the input by
2:37:35
the input by to get the answers out right and that's
2:37:37
to get the answers out right and that's
2:37:37
to get the answers out right and that's a lot of numbers and a lot of functions
2:37:39
a lot of numbers and a lot of functions
2:37:39
a lot of numbers and a lot of functions but it's still just numbers yeah so
2:37:43
but it's still just numbers yeah so
2:37:43
but it's still just numbers yeah so um speaking of noel maybe she can
2:37:46
um speaking of noel maybe she can
2:37:46
um speaking of noel maybe she can just she can help us out here um i was
2:37:49
just she can help us out here um i was
2:37:49
just she can help us out here um i was wondering when we talk about deep
2:37:51
wondering when we talk about deep
2:37:51
wondering when we talk about deep learning
2:37:51
learning
2:37:51
learning um what would be a good good sort of
2:37:55
um what would be a good good sort of
2:37:55
um what would be a good good sort of explainer that we can apply
2:37:56
explainer that we can apply
2:37:56
explainer that we can apply to our problem and and get a sense of
2:37:59
to our problem and and get a sense of
2:37:59
to our problem and and get a sense of what the neural network is doing
2:38:01
what the neural network is doing
2:38:01
what the neural network is doing in the case of predicting tacos versus
2:38:03
in the case of predicting tacos versus
2:38:03
in the case of predicting tacos versus burritos for example
2:38:06
burritos for example
2:38:06
burritos for example yeah that's a great question actually um
2:38:08
yeah that's a great question actually um
2:38:08
yeah that's a great question actually um so there's a few
2:38:10
so there's a few
2:38:10
so there's a few different ones that i would recommend
2:38:13
different ones that i would recommend
2:38:13
different ones that i would recommend um the first one and actually i hope i
2:38:15
um the first one and actually i hope i
2:38:15
um the first one and actually i hope i don't know if everyone's gonna get the
2:38:16
don't know if everyone's gonna get the
2:38:16
don't know if everyone's gonna get the slide deck but i put a few of these
2:38:18
slide deck but i put a few of these
2:38:18
slide deck but i put a few of these tools
2:38:19
tools
2:38:19
tools in our um in the slide deck so i don't
2:38:22
in our um in the slide deck so i don't
2:38:22
in our um in the slide deck so i don't know how that gets shared or if
2:38:23
know how that gets shared or if
2:38:23
know how that gets shared or if people want to just find it yeah we can
2:38:26
people want to just find it yeah we can
2:38:26
people want to just find it yeah we can put it on the website if you want uh
2:38:27
put it on the website if you want uh
2:38:28
put it on the website if you want uh that makes it yeah sure so the first one
2:38:29
that makes it yeah sure so the first one
2:38:29
that makes it yeah sure so the first one is um fair learn is one uh and there's
2:38:33
is um fair learn is one uh and there's
2:38:33
is um fair learn is one uh and there's a really great github-based tutorial
2:38:35
a really great github-based tutorial
2:38:35
a really great github-based tutorial that you can walk through it gives you a
2:38:37
that you can walk through it gives you a
2:38:37
that you can walk through it gives you a sample model
2:38:38
sample model
2:38:38
sample model um there's also interpret ml uh both of
2:38:41
um there's also interpret ml uh both of
2:38:41
um there's also interpret ml uh both of those
2:38:42
those
2:38:42
those give you the ability to just even grab a
2:38:44
give you the ability to just even grab a
2:38:44
give you the ability to just even grab a baseline
2:38:45
baseline
2:38:45
baseline of how interpretable your models are
2:38:48
of how interpretable your models are
2:38:48
of how interpretable your models are and to identify kind of the level of
2:38:51
and to identify kind of the level of
2:38:51
and to identify kind of the level of opaqueness or black
2:38:53
opaqueness or black
2:38:53
opaqueness or black boxiness if you will of your model
2:38:56
boxiness if you will of your model
2:38:56
boxiness if you will of your model yeah that makes sense yeah we've
2:38:58
yeah that makes sense yeah we've
2:38:58
yeah that makes sense yeah we've actually written a lot of stuff about it
2:39:00
actually written a lot of stuff about it
2:39:00
actually written a lot of stuff about it uh uh on our blogs on medium uh so if
2:39:04
uh uh on our blogs on medium uh so if
2:39:04
uh uh on our blogs on medium uh so if people are interested they could
2:39:05
people are interested they could
2:39:05
people are interested they could definitely check those
2:39:06
definitely check those
2:39:06
definitely check those so yeah
2:39:10
so yeah
2:39:10
so yeah it's really interesting but i think also
2:39:12
it's really interesting but i think also
2:39:12
it's really interesting but i think also a lot of it comes down to just the
2:39:13
a lot of it comes down to just the
2:39:14
a lot of it comes down to just the conversations that
2:39:15
conversations that
2:39:15
conversations that um are being had at the beginning of a
2:39:18
um are being had at the beginning of a
2:39:18
um are being had at the beginning of a project and
2:39:19
project and
2:39:19
project and the support of even engineering
2:39:21
the support of even engineering
2:39:21
the support of even engineering management and engineering leadership
2:39:23
management and engineering leadership
2:39:23
management and engineering leadership in pausing because at least in you know
2:39:26
in pausing because at least in you know
2:39:26
in pausing because at least in you know i've been in microsoft ai teams and in
2:39:28
i've been in microsoft ai teams and in
2:39:28
i've been in microsoft ai teams and in amazon ai teams
2:39:30
amazon ai teams
2:39:30
amazon ai teams and in both of those those teams our
2:39:32
and in both of those those teams our
2:39:32
and in both of those those teams our velocity was so high
2:39:34
velocity was so high
2:39:34
velocity was so high that we really didn't it was kind of
2:39:36
that we really didn't it was kind of
2:39:36
that we really didn't it was kind of like you know
2:39:38
like you know
2:39:38
like you know seth's you know message that it was like
2:39:40
seth's you know message that it was like
2:39:40
seth's you know message that it was like this fire hose of
2:39:42
this fire hose of
2:39:42
this fire hose of important things that we were doing and
2:39:44
important things that we were doing and
2:39:44
important things that we were doing and really trying to figure out okay when do
2:39:46
really trying to figure out okay when do
2:39:46
really trying to figure out okay when do we stop and
2:39:47
we stop and
2:39:47
we stop and ask these questions and we didn't
2:39:49
ask these questions and we didn't
2:39:49
ask these questions and we didn't honestly like during those early days of
2:39:50
honestly like during those early days of
2:39:50
honestly like during those early days of alexa when we were
2:39:52
alexa when we were
2:39:52
alexa when we were gaining hundreds of thousands of users a
2:39:54
gaining hundreds of thousands of users a
2:39:54
gaining hundreds of thousands of users a day
2:39:55
day
2:39:55
day like we weren't asking those questions
2:39:57
like we weren't asking those questions
2:39:57
like we weren't asking those questions so we are i mean they are now
2:39:59
so we are i mean they are now
2:39:59
so we are i mean they are now uh but this is what we're all facing we
2:40:02
uh but this is what we're all facing we
2:40:02
uh but this is what we're all facing we don't know if the ai we're building now
2:40:05
don't know if the ai we're building now
2:40:05
don't know if the ai we're building now is the next you know facebook or alexa
2:40:08
is the next you know facebook or alexa
2:40:08
is the next you know facebook or alexa or google search right we don't know so
2:40:10
or google search right we don't know so
2:40:10
or google search right we don't know so you have to ask these questions
2:40:13
you have to ask these questions
2:40:13
you have to ask these questions in the you know as a result of that that
2:40:15
in the you know as a result of that that
2:40:15
in the you know as a result of that that you don't know what you're building in
2:40:17
you don't know what you're building in
2:40:17
you don't know what you're building in ai we don't know the full extent of how
2:40:19
ai we don't know the full extent of how
2:40:19
ai we don't know the full extent of how these technologies will be used
2:40:21
these technologies will be used
2:40:21
these technologies will be used so i just say make it a habit do it
2:40:23
so i just say make it a habit do it
2:40:23
so i just say make it a habit do it anyway i think seth ended it well
2:40:25
anyway i think seth ended it well
2:40:25
anyway i think seth ended it well you know ask the one question um i won't
2:40:28
you know ask the one question um i won't
2:40:28
you know ask the one question um i won't make it about burritos or pizzas but
2:40:30
make it about burritos or pizzas but
2:40:30
make it about burritos or pizzas but that's like ask the question make sure
2:40:33
that's like ask the question make sure
2:40:33
that's like ask the question make sure you know who you're serving
2:40:35
you know who you're serving
2:40:35
you know who you're serving and whether or not these predictions
2:40:36
and whether or not these predictions
2:40:36
and whether or not these predictions actually serve them
2:40:38
actually serve them
2:40:38
actually serve them and i think that's the best way to start
2:40:40
and i think that's the best way to start
2:40:40
and i think that's the best way to start any project and
2:40:41
any project and
2:40:41
any project and and so um we've had some discussion
2:40:44
and so um we've had some discussion
2:40:44
and so um we've had some discussion about a metal ops
2:40:46
about a metal ops
2:40:46
about a metal ops in one of the previous episodes um
2:40:50
in one of the previous episodes um
2:40:50
in one of the previous episodes um what do you think um uh in in terms of a
2:40:53
what do you think um uh in in terms of a
2:40:53
what do you think um uh in in terms of a good approach to
2:40:55
good approach to
2:40:55
good approach to integrating explainers into interpreters
2:40:57
integrating explainers into interpreters
2:40:58
integrating explainers into interpreters into your project
2:40:58
into your project
2:40:58
into your project where would you actually place them in
2:41:01
where would you actually place them in
2:41:01
where would you actually place them in your whole ml
2:41:02
your whole ml
2:41:02
your whole ml ops process yeah so i actually think
2:41:04
ops process yeah so i actually think
2:41:04
ops process yeah so i actually think they belong in a couple different places
2:41:06
they belong in a couple different places
2:41:06
they belong in a couple different places um because there's different impact
2:41:09
um because there's different impact
2:41:09
um because there's different impact zones
2:41:10
zones
2:41:10
zones i call them for explainable models so i
2:41:13
i call them for explainable models so i
2:41:13
i call them for explainable models so i have and i
2:41:15
have and i
2:41:15
have and i i'm trying to think if i have it ready
2:41:16
i'm trying to think if i have it ready
2:41:16
i'm trying to think if i have it ready to show you but let's if we can vision
2:41:18
to show you but let's if we can vision
2:41:18
to show you but let's if we can vision in our mind's eye
2:41:20
in our mind's eye
2:41:20
in our mind's eye um an architecture diagram of like our a
2:41:23
um an architecture diagram of like our a
2:41:23
um an architecture diagram of like our a model development process right and you
2:41:25
model development process right and you
2:41:25
model development process right and you think about
2:41:26
think about
2:41:26
think about five classic stages of building a model
2:41:28
five classic stages of building a model
2:41:28
five classic stages of building a model from
2:41:29
from
2:41:29
from data collection through ingestion
2:41:31
data collection through ingestion
2:41:32
data collection through ingestion training
2:41:32
training
2:41:32
training deployment right if we think about it
2:41:34
deployment right if we think about it
2:41:34
deployment right if we think about it from that perspective
2:41:36
from that perspective
2:41:36
from that perspective there are opportunities for it not just
2:41:39
there are opportunities for it not just
2:41:39
there are opportunities for it not just the tools so i
2:41:40
the tools so i
2:41:40
the tools so i you're kind of you're asking me one
2:41:41
you're kind of you're asking me one
2:41:41
you're kind of you're asking me one question i'm gonna answer it with both
2:41:43
question i'm gonna answer it with both
2:41:44
question i'm gonna answer it with both so one is at the very beginning during
2:41:46
so one is at the very beginning during
2:41:46
so one is at the very beginning during data collection diversity and data
2:41:47
data collection diversity and data
2:41:48
data collection diversity and data collection has to be
2:41:49
collection has to be
2:41:49
collection has to be identified and monitored at the very
2:41:51
identified and monitored at the very
2:41:51
identified and monitored at the very beginning of that process
2:41:53
beginning of that process
2:41:53
beginning of that process what we're now doing because most of us
2:41:55
what we're now doing because most of us
2:41:55
what we're now doing because most of us are dealing with models that have
2:41:56
are dealing with models that have
2:41:56
are dealing with models that have already been trained on some
2:41:58
already been trained on some
2:41:58
already been trained on some form of data and so the next phase of
2:42:00
form of data and so the next phase of
2:42:00
form of data and so the next phase of that is in the actual model pipeline
2:42:03
that is in the actual model pipeline
2:42:03
that is in the actual model pipeline itself
2:42:03
itself
2:42:03
itself so i am a huge fan of mlaps is somewhat
2:42:07
so i am a huge fan of mlaps is somewhat
2:42:07
so i am a huge fan of mlaps is somewhat of a new term probably to a lot of
2:42:08
of a new term probably to a lot of
2:42:08
of a new term probably to a lot of people who are listening
2:42:10
people who are listening
2:42:10
people who are listening as well but building an actual code
2:42:12
as well but building an actual code
2:42:12
as well but building an actual code pipeline for
2:42:13
pipeline for
2:42:14
pipeline for ai i mean again in my early stages of
2:42:16
ai i mean again in my early stages of
2:42:16
ai i mean again in my early stages of alexa
2:42:17
alexa
2:42:17
alexa we had no pipeline there was no pipeline
2:42:19
we had no pipeline there was no pipeline
2:42:19
we had no pipeline there was no pipeline it was just like
2:42:21
it was just like
2:42:21
it was just like deploy like fingers on a keyboard deploy
2:42:24
deploy like fingers on a keyboard deploy
2:42:24
deploy like fingers on a keyboard deploy um
2:42:24
um
2:42:24
um and so i was shocked to find you know
2:42:28
and so i was shocked to find you know
2:42:28
and so i was shocked to find you know like
2:42:28
like
2:42:28
like and i was coming similar to anthony i
2:42:31
and i was coming similar to anthony i
2:42:31
and i was coming similar to anthony i was coming from
2:42:32
was coming from
2:42:32
was coming from a like aws kind of infrastructure
2:42:36
a like aws kind of infrastructure
2:42:36
a like aws kind of infrastructure cloud infrastructure kind of role and i
2:42:38
cloud infrastructure kind of role and i
2:42:38
cloud infrastructure kind of role and i was like why wouldn't we just deploy
2:42:40
was like why wouldn't we just deploy
2:42:40
was like why wouldn't we just deploy this like
2:42:40
this like
2:42:40
this like any other code and we do have like mod
2:42:44
any other code and we do have like mod
2:42:44
any other code and we do have like mod we do have tools that we use to protect
2:42:46
we do have tools that we use to protect
2:42:46
we do have tools that we use to protect the integrity of code
2:42:48
the integrity of code
2:42:48
the integrity of code and now we're trying to apply them to
2:42:49
and now we're trying to apply them to
2:42:49
and now we're trying to apply them to things that have always sat in research
2:42:51
things that have always sat in research
2:42:51
things that have always sat in research so i do think that in the um in the
2:42:53
so i do think that in the um in the
2:42:53
so i do think that in the um in the phase of actual
2:42:54
phase of actual
2:42:54
phase of actual model deployment but also in the phase
2:42:57
model deployment but also in the phase
2:42:57
model deployment but also in the phase of
2:42:58
of
2:42:58
of model creation that we can leverage
2:43:00
model creation that we can leverage
2:43:00
model creation that we can leverage these packages that are are now being
2:43:01
these packages that are are now being
2:43:01
these packages that are are now being made readily accessible
2:43:03
made readily accessible
2:43:03
made readily accessible so you're back from do you want to add
2:43:05
so you're back from do you want to add
2:43:05
so you're back from do you want to add in
2:43:07
in
2:43:07
in i didn't hear that my internet went out
2:43:09
i didn't hear that my internet went out
2:43:09
i didn't hear that my internet went out and so
2:43:10
and so
2:43:10
and so it's terrible terrible right now and
2:43:13
it's terrible terrible right now and
2:43:13
it's terrible terrible right now and half the
2:43:13
half the
2:43:13
half the like a bunch of these countries in the
2:43:15
like a bunch of these countries in the
2:43:15
like a bunch of these countries in the south are like
2:43:16
south are like
2:43:16
south are like handling a hurricane so it's kind of sad
2:43:20
handling a hurricane so it's kind of sad
2:43:20
handling a hurricane so it's kind of sad so what was the question i wanna i'm
2:43:22
so what was the question i wanna i'm
2:43:22
so what was the question i wanna i'm like miss like the all the important
2:43:24
like miss like the all the important
2:43:24
like miss like the all the important stuff and now i get to come in and be
2:43:25
stuff and now i get to come in and be
2:43:25
stuff and now i get to come in and be like
2:43:26
like
2:43:26
like it's good for those who need it also
2:43:28
it's good for those who need it also
2:43:28
it's good for those who need it also repeat it so
2:43:30
repeat it so
2:43:30
repeat it so so in in general you're saying actually
2:43:32
so in in general you're saying actually
2:43:32
so in in general you're saying actually noelle that
2:43:33
noelle that
2:43:33
noelle that that machine learning might not be that
2:43:36
that machine learning might not be that
2:43:36
that machine learning might not be that different from
2:43:37
different from
2:43:37
different from regular software engineering it does uh
2:43:40
regular software engineering it does uh
2:43:40
regular software engineering it does uh get a lot of profit from having these
2:43:42
get a lot of profit from having these
2:43:42
get a lot of profit from having these standard practices of building a
2:43:44
standard practices of building a
2:43:44
standard practices of building a coded pipeline to bring it to production
2:43:47
coded pipeline to bring it to production
2:43:47
coded pipeline to bring it to production so that yeah
2:43:49
so that yeah
2:43:50
so that yeah specifically around pipelines like why
2:43:52
specifically around pipelines like why
2:43:52
specifically around pipelines like why wouldn't we leverage the values that we
2:43:54
wouldn't we leverage the values that we
2:43:54
wouldn't we leverage the values that we see in code pipelines
2:43:55
see in code pipelines
2:43:56
see in code pipelines in deploying which we're now seeing in
2:43:57
in deploying which we're now seeing in
2:43:57
in deploying which we're now seeing in mlx but
2:43:59
mlx but
2:43:59
mlx but like why wouldn't we do that um and that
2:44:01
like why wouldn't we do that um and that
2:44:01
like why wouldn't we do that um and that now if we create a pipeline
2:44:03
now if we create a pipeline
2:44:03
now if we create a pipeline we have different stages of deploying
2:44:05
we have different stages of deploying
2:44:05
we have different stages of deploying that pipeline that we can inject
2:44:07
that pipeline that we can inject
2:44:07
that pipeline that we can inject interpret you know interpretability or
2:44:09
interpret you know interpretability or
2:44:09
interpret you know interpretability or fairness
2:44:10
fairness
2:44:10
fairness um checkers if you will or detectors um
2:44:13
um checkers if you will or detectors um
2:44:13
um checkers if you will or detectors um and there's lots of different packages
2:44:14
and there's lots of different packages
2:44:14
and there's lots of different packages that support that
2:44:15
that support that
2:44:15
that support that across that pipeline i'm almost thinking
2:44:18
across that pipeline i'm almost thinking
2:44:18
across that pipeline i'm almost thinking that this is just a different sort of
2:44:20
that this is just a different sort of
2:44:20
that this is just a different sort of test that you run just like you would
2:44:23
test that you run just like you would
2:44:23
test that you run just like you would run unit tests on your code
2:44:25
run unit tests on your code
2:44:25
run unit tests on your code this could be another test actually
2:44:28
this could be another test actually
2:44:28
this could be another test actually yeah i do think we're a little it's soon
2:44:31
yeah i do think we're a little it's soon
2:44:31
yeah i do think we're a little it's soon um so i don't think we've got the
2:44:32
um so i don't think we've got the
2:44:32
um so i don't think we've got the maturity in those types of tests
2:44:35
maturity in those types of tests
2:44:35
maturity in those types of tests and the results that they give because
2:44:37
and the results that they give because
2:44:37
and the results that they give because right now the results that come out a
2:44:38
right now the results that come out a
2:44:38
right now the results that come out a lot of these tests
2:44:40
lot of these tests
2:44:40
lot of these tests have to be discerned by a human and then
2:44:43
have to be discerned by a human and then
2:44:43
have to be discerned by a human and then we have to go and fix them and you know
2:44:46
we have to go and fix them and you know
2:44:46
we have to go and fix them and you know just like
2:44:46
just like
2:44:46
just like automated ml or some of our automated ai
2:44:49
automated ml or some of our automated ai
2:44:49
automated ml or some of our automated ai tools eventually we can use ai to fix
2:44:52
tools eventually we can use ai to fix
2:44:52
tools eventually we can use ai to fix the problems that it detects
2:44:53
the problems that it detects
2:44:53
the problems that it detects in itself but we're not there yet
2:44:56
in itself but we're not there yet
2:44:56
in itself but we're not there yet especially in explainability
2:44:58
especially in explainability
2:44:58
especially in explainability oh yeah i would add like noel is a
2:45:01
oh yeah i would add like noel is a
2:45:01
oh yeah i would add like noel is a thousand percent
2:45:01
thousand percent
2:45:01
thousand percent right we for some reason in general
2:45:05
right we for some reason in general
2:45:05
right we for some reason in general data science is like because machine
2:45:07
data science is like because machine
2:45:07
data science is like because machine learning models are now going into
2:45:09
learning models are now going into
2:45:09
learning models are now going into production
2:45:10
production
2:45:10
production it's like 1995 all over again for
2:45:12
it's like 1995 all over again for
2:45:12
it's like 1995 all over again for programmers remember in programming
2:45:13
programmers remember in programming
2:45:14
programmers remember in programming we used to like oh we should probably
2:45:15
we used to like oh we should probably
2:45:15
we used to like oh we should probably use source control
2:45:17
use source control
2:45:17
use source control right like right now well you know this
2:45:19
right like right now well you know this
2:45:19
right like right now well you know this better than anyone people are just
2:45:21
better than anyone people are just
2:45:21
better than anyone people are just emailing models around
2:45:23
emailing models around
2:45:23
emailing models around like we used to email code around and so
2:45:25
like we used to email code around and so
2:45:25
like we used to email code around and so now we're getting into robust
2:45:27
now we're getting into robust
2:45:27
now we're getting into robust software practice because after all it's
2:45:29
software practice because after all it's
2:45:29
software practice because after all it's just software
2:45:30
just software
2:45:30
just software and when it comes to like here's a
2:45:31
and when it comes to like here's a
2:45:31
and when it comes to like here's a simple thing maybe you should have
2:45:34
simple thing maybe you should have
2:45:34
simple thing maybe you should have some examples of things that it always
2:45:36
some examples of things that it always
2:45:36
some examples of things that it always needs to be non-biased at
2:45:38
needs to be non-biased at
2:45:38
needs to be non-biased at put them in unit tests in the pipeline
2:45:40
put them in unit tests in the pipeline
2:45:40
put them in unit tests in the pipeline and say here's a picture of
2:45:42
and say here's a picture of
2:45:42
and say here's a picture of a pizza and it should guess none right
2:45:45
a pizza and it should guess none right
2:45:45
a pizza and it should guess none right maybe just put that as a unit test
2:45:46
maybe just put that as a unit test
2:45:46
maybe just put that as a unit test as part of the build process maybe maybe
2:45:49
as part of the build process maybe maybe
2:45:49
as part of the build process maybe maybe cr
2:45:49
cr
2:45:49
cr maybe have images or maybe have examples
2:45:51
maybe have images or maybe have examples
2:45:52
maybe have images or maybe have examples of of things that it should not be doing
2:45:54
of of things that it should not be doing
2:45:54
of of things that it should not be doing and have it
2:45:54
and have it
2:45:54
and have it put it in unit test and fail it and
2:45:56
put it in unit test and fail it and
2:45:56
put it in unit test and fail it and that's like a really
2:45:58
that's like a really
2:45:58
that's like a really really good start but noelle is right
2:45:59
really good start but noelle is right
2:45:59
really good start but noelle is right the the the
2:46:01
the the the
2:46:01
the the the important responsible ml stuff
2:46:04
important responsible ml stuff
2:46:04
important responsible ml stuff like fair learn for example the white
2:46:06
like fair learn for example the white
2:46:06
like fair learn for example the white box models that look
2:46:08
box models that look
2:46:08
box models that look inside of decision trees for example or
2:46:10
inside of decision trees for example or
2:46:10
inside of decision trees for example or the black box models
2:46:11
the black box models
2:46:11
the black box models that as noel actually pointed out swap
2:46:14
that as noel actually pointed out swap
2:46:14
that as noel actually pointed out swap features in and out to see
2:46:16
features in and out to see
2:46:16
features in and out to see like how much does it change like here's
2:46:17
like how much does it change like here's
2:46:17
like how much does it change like here's a simple example if you change
2:46:19
a simple example if you change
2:46:19
a simple example if you change if you if gender is one of your features
2:46:21
if you if gender is one of your features
2:46:21
if you if gender is one of your features and you take it out
2:46:22
and you take it out
2:46:22
and you take it out and all of a sudden it starts predicting
2:46:24
and all of a sudden it starts predicting
2:46:24
and all of a sudden it starts predicting differently or the disparity between
2:46:26
differently or the disparity between
2:46:26
differently or the disparity between genders is huge
2:46:27
genders is huge
2:46:27
genders is huge you have a problem you could test that
2:46:29
you have a problem you could test that
2:46:29
you have a problem you could test that pretty easily
2:46:30
pretty easily
2:46:30
pretty easily right and you should and so i'm a
2:46:32
right and you should and so i'm a
2:46:32
right and you should and so i'm a thousand percent in agreement with what
2:46:33
thousand percent in agreement with what
2:46:33
thousand percent in agreement with what noel is saying
2:46:34
noel is saying
2:46:34
noel is saying no and and um
2:46:38
no and and um
2:46:38
no and and um so we we've talked about these pipelines
2:46:40
so we we've talked about these pipelines
2:46:40
so we we've talked about these pipelines and this is sort of the quality check
2:46:42
and this is sort of the quality check
2:46:42
and this is sort of the quality check going into production
2:46:43
going into production
2:46:43
going into production and i'm thinking also once you you are
2:46:46
and i'm thinking also once you you are
2:46:46
and i'm thinking also once you you are in production
2:46:47
in production
2:46:47
in production things might change actually so you
2:46:49
things might change actually so you
2:46:49
things might change actually so you might run into problems where maybe
2:46:51
might run into problems where maybe
2:46:51
might run into problems where maybe the samples in production change from
2:46:53
the samples in production change from
2:46:53
the samples in production change from the samples that we've used for training
2:46:55
the samples that we've used for training
2:46:55
the samples that we've used for training is there anything we can do
2:46:56
is there anything we can do
2:46:56
is there anything we can do about that actually yeah
2:47:01
about that actually yeah
2:47:01
about that actually yeah okay no please you know l
2:47:04
okay no please you know l
2:47:04
okay no please you know l i was just gonna say that one thing i
2:47:06
i was just gonna say that one thing i
2:47:06
i was just gonna say that one thing i always try to encourage
2:47:07
always try to encourage
2:47:07
always try to encourage organizations is to not think about
2:47:09
organizations is to not think about
2:47:09
organizations is to not think about production in the same way like
2:47:11
production in the same way like
2:47:11
production in the same way like ai is the definition of a perpetual beta
2:47:15
ai is the definition of a perpetual beta
2:47:15
ai is the definition of a perpetual beta so i don't feel like there's the classic
2:47:18
so i don't feel like there's the classic
2:47:18
so i don't feel like there's the classic you know sense of like pristine
2:47:20
you know sense of like pristine
2:47:20
you know sense of like pristine production which again is just a
2:47:22
production which again is just a
2:47:22
production which again is just a paradigm shift
2:47:23
paradigm shift
2:47:23
paradigm shift from kind of a software engineering
2:47:25
from kind of a software engineering
2:47:25
from kind of a software engineering world where we push something in
2:47:26
world where we push something in
2:47:26
world where we push something in production and it
2:47:28
production and it
2:47:28
production and it and it's done where an ai model we would
2:47:31
and it's done where an ai model we would
2:47:31
and it's done where an ai model we would i mean we were changing you know alexa
2:47:33
i mean we were changing you know alexa
2:47:34
i mean we were changing you know alexa and even now
2:47:35
and even now
2:47:35
and even now at microsoft and cognitive services like
2:47:37
at microsoft and cognitive services like
2:47:37
at microsoft and cognitive services like those models are being updated on the
2:47:38
those models are being updated on the
2:47:38
those models are being updated on the fly
2:47:39
fly
2:47:39
fly your decisions in a pre-built model can
2:47:42
your decisions in a pre-built model can
2:47:42
your decisions in a pre-built model can change
2:47:43
change
2:47:43
change the next time you call it um which if
2:47:45
the next time you call it um which if
2:47:45
the next time you call it um which if you guys
2:47:46
you guys
2:47:46
you guys if you all remember in the early days of
2:47:47
if you all remember in the early days of
2:47:47
if you all remember in the early days of alexa you could ask it a question
2:47:49
alexa you could ask it a question
2:47:49
alexa you could ask it a question and the very next day ask it the same
2:47:51
and the very next day ask it the same
2:47:51
and the very next day ask it the same question and the answer would be
2:47:52
question and the answer would be
2:47:52
question and the answer would be different because we were constantly
2:47:54
different because we were constantly
2:47:54
different because we were constantly updating it so i think i always
2:47:56
updating it so i think i always
2:47:56
updating it so i think i always encourage people to
2:47:58
encourage people to
2:47:58
encourage people to remember these are experiments that we
2:48:00
remember these are experiments that we
2:48:00
remember these are experiments that we are choosing to put in production
2:48:02
are choosing to put in production
2:48:02
are choosing to put in production and those experiments don't change just
2:48:04
and those experiments don't change just
2:48:04
and those experiments don't change just due to the nature of their
2:48:05
due to the nature of their
2:48:05
due to the nature of their deployment some mathematical things you
2:48:08
deployment some mathematical things you
2:48:08
deployment some mathematical things you can do just to follow on with what
2:48:10
can do just to follow on with what
2:48:10
can do just to follow on with what noel is saying is you have the original
2:48:12
noel is saying is you have the original
2:48:12
noel is saying is you have the original data set you use to train especially if
2:48:14
data set you use to train especially if
2:48:14
data set you use to train especially if you have
2:48:15
you have
2:48:15
you have what's called data providence through
2:48:16
what's called data providence through
2:48:16
what's called data providence through your models you know what data was used
2:48:18
your models you know what data was used
2:48:18
your models you know what data was used to train it
2:48:19
to train it
2:48:19
to train it you should also store the data that
2:48:22
you should also store the data that
2:48:22
you should also store the data that you're doing what's called
2:48:23
you're doing what's called
2:48:23
you're doing what's called inference on which is the prediction
2:48:25
inference on which is the prediction
2:48:25
inference on which is the prediction right and what you can do
2:48:26
right and what you can do
2:48:26
right and what you can do is you can use unsupervised machine
2:48:28
is you can use unsupervised machine
2:48:28
is you can use unsupervised machine learning to actually detect
2:48:30
learning to actually detect
2:48:30
learning to actually detect if there is a large difference between
2:48:33
if there is a large difference between
2:48:33
if there is a large difference between the data set you use
2:48:35
the data set you use
2:48:35
the data set you use to train your your actual uh
2:48:38
to train your your actual uh
2:48:38
to train your your actual uh model and the data set that you're
2:48:39
model and the data set that you're
2:48:39
model and the data set that you're actually doing inference on this is
2:48:41
actually doing inference on this is
2:48:41
actually doing inference on this is called data drift or sometimes
2:48:43
called data drift or sometimes
2:48:43
called data drift or sometimes model drip and you can create a machine
2:48:45
model drip and you can create a machine
2:48:45
model drip and you can create a machine learning model
2:48:47
learning model
2:48:47
learning model that gets the appropriate tolerance for
2:48:50
that gets the appropriate tolerance for
2:48:50
that gets the appropriate tolerance for change
2:48:50
change
2:48:50
change obviously this is this is more of a
2:48:52
obviously this is this is more of a
2:48:52
obviously this is this is more of a robust scenario people that have been
2:48:54
robust scenario people that have been
2:48:54
robust scenario people that have been doing an ai
2:48:55
doing an ai
2:48:55
doing an ai or machine learning for a long time but
2:48:57
or machine learning for a long time but
2:48:57
or machine learning for a long time but you can actually measure
2:48:58
you can actually measure
2:48:58
you can actually measure the difference in the data sets in using
2:49:01
the difference in the data sets in using
2:49:02
the difference in the data sets in using unsupervised line for example you could
2:49:03
unsupervised line for example you could
2:49:03
unsupervised line for example you could do
2:49:03
do
2:49:03
do you could do on your original data use a
2:49:05
you could do on your original data use a
2:49:05
you could do on your original data use a train you could do k-means
2:49:07
train you could do k-means
2:49:07
train you could do k-means and find the centers of the data right
2:49:08
and find the centers of the data right
2:49:08
and find the centers of the data right and then output that into an
2:49:10
and then output that into an
2:49:10
and then output that into an n-dimensional vector
2:49:11
n-dimensional vector
2:49:11
n-dimensional vector you could do the same as soon as you
2:49:13
you could do the same as soon as you
2:49:13
you could do the same as soon as you start collecting a lot of data into an
2:49:15
start collecting a lot of data into an
2:49:15
start collecting a lot of data into an n-dimensional vector and then you can
2:49:17
n-dimensional vector and then you can
2:49:17
n-dimensional vector and then you can measure the distance
2:49:19
measure the distance
2:49:19
measure the distance and if the distance is super large you
2:49:21
and if the distance is super large you
2:49:21
and if the distance is super large you can say um we need to retrain
2:49:23
can say um we need to retrain
2:49:23
can say um we need to retrain but as noelle said that still never will
2:49:26
but as noelle said that still never will
2:49:26
but as noelle said that still never will take away
2:49:27
take away
2:49:27
take away from the human aspect of if the data is
2:49:30
from the human aspect of if the data is
2:49:30
from the human aspect of if the data is biased
2:49:31
biased
2:49:31
biased even during inference it's still going
2:49:34
even during inference it's still going
2:49:34
even during inference it's still going to have a problem right even if the
2:49:35
to have a problem right even if the
2:49:36
to have a problem right even if the distance between the unsupervised
2:49:38
distance between the unsupervised
2:49:38
distance between the unsupervised data between the unsupervised models
2:49:40
data between the unsupervised models
2:49:40
data between the unsupervised models between the data like the
2:49:41
between the data like the
2:49:42
between the data like the the train data and the inference data
2:49:43
the train data and the inference data
2:49:43
the train data and the inference data even if the the difference is small
2:49:45
even if the the difference is small
2:49:46
even if the the difference is small you will still miss certain things
2:49:48
you will still miss certain things
2:49:48
you will still miss certain things especially
2:49:49
especially
2:49:50
especially if your model is still biased right and
2:49:52
if your model is still biased right and
2:49:52
if your model is still biased right and that that's a problem that will not be
2:49:54
that that's a problem that will not be
2:49:54
that that's a problem that will not be solved
2:49:54
solved
2:49:54
solved with those empirical methods but it's
2:49:56
with those empirical methods but it's
2:49:56
with those empirical methods but it's it's a it's a check
2:49:57
it's a it's a check
2:49:57
it's a it's a check that you could use i think it's a sort
2:50:00
that you could use i think it's a sort
2:50:00
that you could use i think it's a sort of an additional safety net and we can't
2:50:02
of an additional safety net and we can't
2:50:02
of an additional safety net and we can't have enough of those
2:50:08
it all yeah yeah i mean
2:50:11
it all yeah yeah i mean
2:50:12
it all yeah yeah i mean and and this is some of the more
2:50:14
and and this is some of the more
2:50:14
and and this is some of the more advanced stuff i
2:50:15
advanced stuff i
2:50:15
advanced stuff i i guess that there's still a lot of
2:50:17
i guess that there's still a lot of
2:50:17
i guess that there's still a lot of people that are getting into
2:50:18
people that are getting into
2:50:18
people that are getting into the ai space and trying to make sense of
2:50:21
the ai space and trying to make sense of
2:50:21
the ai space and trying to make sense of everything that they need to do
2:50:22
everything that they need to do
2:50:22
everything that they need to do and this sounds like a a lot of work
2:50:25
and this sounds like a a lot of work
2:50:26
and this sounds like a a lot of work actually
2:50:26
actually
2:50:26
actually if i'm honest so i'm thinking what what
2:50:29
if i'm honest so i'm thinking what what
2:50:29
if i'm honest so i'm thinking what what would be
2:50:29
would be
2:50:29
would be actually the first thing that you would
2:50:31
actually the first thing that you would
2:50:31
actually the first thing that you would do um noel
2:50:33
do um noel
2:50:33
do um noel when talking about improving the ethical
2:50:36
when talking about improving the ethical
2:50:36
when talking about improving the ethical side of ai
2:50:38
side of ai
2:50:38
side of ai but if we're talking about beginners um
2:50:41
but if we're talking about beginners um
2:50:41
but if we're talking about beginners um i'm
2:50:41
i'm
2:50:41
i'm always telling you i tell ceos that are
2:50:44
always telling you i tell ceos that are
2:50:44
always telling you i tell ceos that are like
2:50:44
like
2:50:44
like investing in ai for the first time my
2:50:47
investing in ai for the first time my
2:50:47
investing in ai for the first time my you have to have a like a strategy we
2:50:50
you have to have a like a strategy we
2:50:50
you have to have a like a strategy we call like an ai manifesto you have to
2:50:52
call like an ai manifesto you have to
2:50:52
call like an ai manifesto you have to have a decision
2:50:53
have a decision
2:50:53
have a decision and you need to align ethics ethical use
2:50:56
and you need to align ethics ethical use
2:50:56
and you need to align ethics ethical use of this technology with your business
2:50:58
of this technology with your business
2:50:58
of this technology with your business values because your business
2:51:00
values because your business
2:51:00
values because your business meaning those who are going to come to
2:51:01
meaning those who are going to come to
2:51:01
meaning those who are going to come to the models and say i have this problem
2:51:03
the models and say i have this problem
2:51:03
the models and say i have this problem and i need it solved
2:51:04
and i need it solved
2:51:04
and i need it solved are going to be incented in certain ways
2:51:07
are going to be incented in certain ways
2:51:07
are going to be incented in certain ways and that incentive needs to
2:51:08
and that incentive needs to
2:51:08
and that incentive needs to include ethical and responsible use of
2:51:11
include ethical and responsible use of
2:51:11
include ethical and responsible use of this technology so i always
2:51:13
this technology so i always
2:51:13
this technology so i always i always encourage people that they need
2:51:14
i always encourage people that they need
2:51:14
i always encourage people that they need to have
2:51:16
to have
2:51:16
to have they need to have the backing of those
2:51:18
they need to have the backing of those
2:51:18
they need to have the backing of those asking the question
2:51:19
asking the question
2:51:19
asking the question because usually as a modeler we're not
2:51:21
because usually as a modeler we're not
2:51:21
because usually as a modeler we're not solving our own
2:51:22
solving our own
2:51:22
solving our own problems we're solving a problem that's
2:51:24
problems we're solving a problem that's
2:51:24
problems we're solving a problem that's been presented to us
2:51:26
been presented to us
2:51:26
been presented to us so we need to make sure that we have
2:51:27
so we need to make sure that we have
2:51:27
so we need to make sure that we have executive support or even just
2:51:30
executive support or even just
2:51:30
executive support or even just managerial support for pulling the chain
2:51:34
managerial support for pulling the chain
2:51:34
managerial support for pulling the chain on the train and stopping if or pausing
2:51:37
on the train and stopping if or pausing
2:51:37
on the train and stopping if or pausing or asking different questions
2:51:39
or asking different questions
2:51:39
or asking different questions if our model starts to behave badly the
2:51:41
if our model starts to behave badly the
2:51:41
if our model starts to behave badly the other thing i would say
2:51:43
other thing i would say
2:51:43
other thing i would say and i'll only say two things is that um
2:51:46
and i'll only say two things is that um
2:51:46
and i'll only say two things is that um also having just making the decision
2:51:49
also having just making the decision
2:51:49
also having just making the decision to leverage at the very least
2:51:52
to leverage at the very least
2:51:52
to leverage at the very least um a fairness package a simple
2:51:56
um a fairness package a simple
2:51:56
um a fairness package a simple line of code that you would execute upon
2:51:59
line of code that you would execute upon
2:52:00
line of code that you would execute upon testing your model just to say like
2:52:02
testing your model just to say like
2:52:02
testing your model just to say like here's
2:52:03
here's
2:52:03
here's how these um confident like here's how
2:52:05
how these um confident like here's how
2:52:06
how these um confident like here's how the confidence
2:52:07
the confidence
2:52:07
the confidence um is performing against our baseline or
2:52:10
um is performing against our baseline or
2:52:10
um is performing against our baseline or here's how like do something i even
2:52:13
here's how like do something i even
2:52:13
here's how like do something i even asked the question force everybody who
2:52:15
asked the question force everybody who
2:52:15
asked the question force everybody who does something with their model to
2:52:16
does something with their model to
2:52:16
does something with their model to explain every decision that they make
2:52:18
explain every decision that they make
2:52:18
explain every decision that they make in a stand-up it doesn't have to be hard
2:52:21
in a stand-up it doesn't have to be hard
2:52:21
in a stand-up it doesn't have to be hard it doesn't have to even be
2:52:22
it doesn't have to even be
2:52:22
it doesn't have to even be in code but it does have to be model
2:52:24
in code but it does have to be model
2:52:24
in code but it does have to be model driven
2:52:26
driven
2:52:26
driven oh doesn't yeah i would not no i would
2:52:29
oh doesn't yeah i would not no i would
2:52:29
oh doesn't yeah i would not no i would also add a simple
2:52:30
also add a simple
2:52:30
also add a simple a simple test like maybe we should as a
2:52:33
a simple test like maybe we should as a
2:52:33
a simple test like maybe we should as a society
2:52:34
society
2:52:34
society decide that we need to add a disclaimer
2:52:37
decide that we need to add a disclaimer
2:52:37
decide that we need to add a disclaimer and say
2:52:38
and say
2:52:38
and say this decision has been aided by
2:52:40
this decision has been aided by
2:52:40
this decision has been aided by mechanical means
2:52:42
mechanical means
2:52:42
mechanical means right just knowing just knowing like i
2:52:45
right just knowing just knowing like i
2:52:45
right just knowing just knowing like i was talking to somebody in
2:52:46
was talking to somebody in
2:52:46
was talking to somebody in a smaller country and they were they
2:52:47
a smaller country and they were they
2:52:47
a smaller country and they were they were trying to come up with some good ai
2:52:50
were trying to come up with some good ai
2:52:50
were trying to come up with some good ai based laws that they might come on like
2:52:52
based laws that they might come on like
2:52:52
based laws that they might come on like hey look a simple one is just say
2:52:54
hey look a simple one is just say
2:52:54
hey look a simple one is just say this decision was partly aided by
2:52:57
this decision was partly aided by
2:52:57
this decision was partly aided by mechanical methods
2:52:59
mechanical methods
2:52:59
mechanical methods right away you can you can start to
2:53:01
right away you can you can start to
2:53:01
right away you can you can start to think about the ethical implications
2:53:04
think about the ethical implications
2:53:04
think about the ethical implications of these things right was i denied alone
2:53:06
of these things right was i denied alone
2:53:06
of these things right was i denied alone because their model is bad or because a
2:53:08
because their model is bad or because a
2:53:08
because their model is bad or because a human looked at it
2:53:09
human looked at it
2:53:09
human looked at it right that and that's a both can still
2:53:11
right that and that's a both can still
2:53:11
right that and that's a both can still be problems right
2:53:12
be problems right
2:53:12
be problems right uh but at least you know there's some
2:53:14
uh but at least you know there's some
2:53:14
uh but at least you know there's some disclosure
2:53:16
disclosure
2:53:16
disclosure going on there but yeah like noel said a
2:53:17
going on there but yeah like noel said a
2:53:17
going on there but yeah like noel said a single test
2:53:19
single test
2:53:19
single test is easy to do and if you as a company
2:53:23
is easy to do and if you as a company
2:53:23
is easy to do and if you as a company uh have an ethos that you do not want to
2:53:25
uh have an ethos that you do not want to
2:53:25
uh have an ethos that you do not want to discriminate based on gender
2:53:27
discriminate based on gender
2:53:27
discriminate based on gender you know nationality color whatever
2:53:29
you know nationality color whatever
2:53:29
you know nationality color whatever maybe you should
2:53:31
maybe you should
2:53:31
maybe you should put those in as tests for your model
2:53:34
put those in as tests for your model
2:53:34
put those in as tests for your model just as simple like invent a person and
2:53:37
just as simple like invent a person and
2:53:37
just as simple like invent a person and make it so
2:53:37
make it so
2:53:38
make it so that it's exactly the thing that you do
2:53:40
that it's exactly the thing that you do
2:53:40
that it's exactly the thing that you do not want to discriminate put it in as a
2:53:42
not want to discriminate put it in as a
2:53:42
not want to discriminate put it in as a single line code uh it's not hard to do
2:53:46
single line code uh it's not hard to do
2:53:46
single line code uh it's not hard to do it's a unit test for fairness and you
2:53:48
it's a unit test for fairness and you
2:53:48
it's a unit test for fairness and you might find that your model is still
2:53:49
might find that your model is still
2:53:50
might find that your model is still unfair
2:53:50
unfair
2:53:50
unfair add another unit test right um but like
2:53:53
add another unit test right um but like
2:53:53
add another unit test right um but like noelle said there's
2:53:54
noelle said there's
2:53:54
noelle said there's just a tiny with a tiny bit of effort
2:53:56
just a tiny with a tiny bit of effort
2:53:56
just a tiny with a tiny bit of effort one line you can start to test
2:53:59
one line you can start to test
2:53:59
one line you can start to test a lot of things like that yeah
2:54:02
a lot of things like that yeah
2:54:02
a lot of things like that yeah yeah it sounds i mean um
2:54:07
yeah it sounds i mean um
2:54:07
yeah it sounds i mean um thinking about it in my in my own
2:54:08
thinking about it in my in my own
2:54:08
thinking about it in my in my own projects um
2:54:10
projects um
2:54:10
projects um it doesn't sound all that complicated to
2:54:13
it doesn't sound all that complicated to
2:54:13
it doesn't sound all that complicated to actually include a few
2:54:14
actually include a few
2:54:14
actually include a few extra unit tests down the line and
2:54:16
extra unit tests down the line and
2:54:16
extra unit tests down the line and perform a few extra steps those are
2:54:18
perform a few extra steps those are
2:54:18
perform a few extra steps those are small enough to uh
2:54:20
small enough to uh
2:54:20
small enough to uh to be done in in seconds actually in
2:54:23
to be done in in seconds actually in
2:54:23
to be done in in seconds actually in most build pipelines
2:54:24
most build pipelines
2:54:24
most build pipelines um so uh yeah especially once you do it
2:54:27
um so uh yeah especially once you do it
2:54:28
um so uh yeah especially once you do it once right like all you have to do is
2:54:29
once right like all you have to do is
2:54:29
once right like all you have to do is ask the question one time build one test
2:54:31
ask the question one time build one test
2:54:31
ask the question one time build one test and it can scale across your whole
2:54:33
and it can scale across your whole
2:54:33
and it can scale across your whole organization um the problem is
2:54:35
organization um the problem is
2:54:35
organization um the problem is oftentimes just like with those of us
2:54:37
oftentimes just like with those of us
2:54:37
oftentimes just like with those of us who have been in software engineering
2:54:39
who have been in software engineering
2:54:39
who have been in software engineering teams
2:54:39
teams
2:54:39
teams that were hesitant to test building the
2:54:42
that were hesitant to test building the
2:54:42
that were hesitant to test building the test is actually
2:54:43
test is actually
2:54:43
test is actually it could be hard it could take a while
2:54:45
it could be hard it could take a while
2:54:45
it could be hard it could take a while to figure out who is that person
2:54:47
to figure out who is that person
2:54:47
to figure out who is that person but it's worth the investment because
2:54:49
but it's worth the investment because
2:54:49
but it's worth the investment because it's a one-time investment and then and
2:54:51
it's a one-time investment and then and
2:54:51
it's a one-time investment and then and as
2:54:51
as
2:54:51
as as all of us who are test driven
2:54:53
as all of us who are test driven
2:54:53
as all of us who are test driven developers know once you build
2:54:55
developers know once you build
2:54:55
developers know once you build successfully against a test you never
2:54:57
successfully against a test you never
2:54:57
successfully against a test you never really want to build any other way
2:54:59
really want to build any other way
2:54:59
really want to build any other way like it it cleans up your code you get
2:55:01
like it it cleans up your code you get
2:55:01
like it it cleans up your code you get 100
2:55:02
100
2:55:02
100 code coverage like all sorts of good
2:55:04
code coverage like all sorts of good
2:55:04
code coverage like all sorts of good things come out of it
2:55:05
things come out of it
2:55:05
things come out of it and so yeah there's a little bit of an
2:55:07
and so yeah there's a little bit of an
2:55:07
and so yeah there's a little bit of an investment
2:55:08
investment
2:55:08
investment um you know it will be not a lot of code
2:55:10
um you know it will be not a lot of code
2:55:10
um you know it will be not a lot of code but understanding what code to write
2:55:12
but understanding what code to write
2:55:12
but understanding what code to write is going to take some questions but
2:55:19
all right oh sounds gone oh oh
2:55:22
all right oh sounds gone oh oh
2:55:22
all right oh sounds gone oh oh can y'all hear me yes i'm sorry i'm here
2:55:26
can y'all hear me yes i'm sorry i'm here
2:55:26
can y'all hear me yes i'm sorry i'm here sorry about it no problem
2:55:37
yeah and i'll just add to what noel is
2:55:39
yeah and i'll just add to what noel is
2:55:39
yeah and i'll just add to what noel is saying basically
2:55:40
saying basically
2:55:40
saying basically model machine learning models are just
2:55:42
model machine learning models are just
2:55:42
model machine learning models are just functions that were lazily written with
2:55:44
functions that were lazily written with
2:55:44
functions that were lazily written with data
2:55:45
data
2:55:45
data so if you think of a model as a function
2:55:48
so if you think of a model as a function
2:55:48
so if you think of a model as a function anything programmers can do with
2:55:49
anything programmers can do with
2:55:49
anything programmers can do with functions
2:55:50
functions
2:55:50
functions you can do with models once they've been
2:55:52
you can do with models once they've been
2:55:52
you can do with models once they've been created you can unit test them you can
2:55:54
created you can unit test them you can
2:55:54
created you can unit test them you can you can email them like i literally
2:55:57
you can email them like i literally
2:55:57
you can email them like i literally showed you the one i showed you is
2:55:58
showed you the one i showed you is
2:55:58
showed you the one i showed you is literally like a
2:56:00
literally like a
2:56:00
literally like a model.onnx it's a file and you i clicked
2:56:03
model.onnx it's a file and you i clicked
2:56:03
model.onnx it's a file and you i clicked on it opened it in netron you can see
2:56:05
on it opened it in netron you can see
2:56:05
on it opened it in netron you can see with
2:56:05
with
2:56:05
with the innards of it there was no rainbows
2:56:07
the innards of it there was no rainbows
2:56:07
the innards of it there was no rainbows and butterflies in there it was just a
2:56:08
and butterflies in there it was just a
2:56:08
and butterflies in there it was just a bunch of numbers it was kind of boring
2:56:10
bunch of numbers it was kind of boring
2:56:10
bunch of numbers it was kind of boring actually blends email
2:56:12
actually blends email
2:56:12
actually blends email models yeah i know i totally
2:56:15
models yeah i know i totally
2:56:15
models yeah i know i totally do people do that that stuff so alicia
2:56:18
do people do that that stuff so alicia
2:56:18
do people do that that stuff so alicia how is that in your work
2:56:19
how is that in your work
2:56:19
how is that in your work um how do you approach this sort of
2:56:22
um how do you approach this sort of
2:56:22
um how do you approach this sort of stuff
2:56:23
stuff
2:56:23
stuff um in your team so um
2:56:26
um in your team so um
2:56:26
um in your team so um we do find that there there is a
2:56:30
we do find that there there is a
2:56:30
we do find that there there is a kind of like a growth cycle with uh data
2:56:33
kind of like a growth cycle with uh data
2:56:33
kind of like a growth cycle with uh data science teams
2:56:34
science teams
2:56:34
science teams so you know depending on the maturity of
2:56:37
so you know depending on the maturity of
2:56:37
so you know depending on the maturity of the data science team and the size of
2:56:38
the data science team and the size of
2:56:38
the data science team and the size of the organization
2:56:40
the organization
2:56:40
the organization um but definitely part of the
2:56:42
um but definitely part of the
2:56:42
um but definitely part of the infrastructure that we
2:56:44
infrastructure that we
2:56:44
infrastructure that we do put in place at a lot of our customer
2:56:47
do put in place at a lot of our customer
2:56:47
do put in place at a lot of our customer sites
2:56:47
sites
2:56:48
sites is the the mlaps for
2:56:51
is the the mlaps for
2:56:51
is the the mlaps for for their data science projects so um
2:56:53
for their data science projects so um
2:56:53
for their data science projects so um and this is part of
2:56:55
and this is part of
2:56:55
and this is part of making the process repeatable and having
2:56:58
making the process repeatable and having
2:56:58
making the process repeatable and having faith
2:56:58
faith
2:56:58
faith in in the recommendations coming out of
2:57:01
in in the recommendations coming out of
2:57:02
in in the recommendations coming out of the
2:57:02
the
2:57:02
the experiments so um yes
2:57:05
experiments so um yes
2:57:06
experiments so um yes they're base function testing
2:57:09
they're base function testing
2:57:09
they're base function testing is is much needed for many of these
2:57:12
is is much needed for many of these
2:57:12
is is much needed for many of these projects
2:57:14
projects
2:57:14
projects and i do see um
2:57:17
and i do see um
2:57:17
and i do see um to this day a huge organizations who
2:57:21
to this day a huge organizations who
2:57:21
to this day a huge organizations who have data scientists who
2:57:22
have data scientists who
2:57:22
have data scientists who who write their code on their laptop and
2:57:25
who write their code on their laptop and
2:57:25
who write their code on their laptop and and they're the only ones
2:57:27
and they're the only ones
2:57:27
and they're the only ones um with access to the experiments so um
2:57:30
um with access to the experiments so um
2:57:30
um with access to the experiments so um a very strong supporter for ml ops
2:57:34
a very strong supporter for ml ops
2:57:34
a very strong supporter for ml ops in the data science space and
2:57:37
in the data science space and
2:57:38
in the data science space and and come to think of it so i've got a
2:57:39
and come to think of it so i've got a
2:57:39
and come to think of it so i've got a lot of data scientists that i work with
2:57:42
lot of data scientists that i work with
2:57:42
lot of data scientists that i work with on teams and one of the things that we
2:57:45
on teams and one of the things that we
2:57:45
on teams and one of the things that we started doing actually
2:57:47
started doing actually
2:57:47
started doing actually at infosupport is we know that there's
2:57:50
at infosupport is we know that there's
2:57:50
at infosupport is we know that there's two kinds of data scientists in the
2:57:52
two kinds of data scientists in the
2:57:52
two kinds of data scientists in the world
2:57:52
world
2:57:52
world roughly there's the da type the
2:57:55
roughly there's the da type the
2:57:55
roughly there's the da type the analytical persons who are
2:57:57
analytical persons who are
2:57:57
analytical persons who are really good at data analysis but not so
2:57:59
really good at data analysis but not so
2:57:59
really good at data analysis but not so good at building
2:58:01
good at building
2:58:01
good at building models using software engineering
2:58:03
models using software engineering
2:58:03
models using software engineering practices the focus has been more on
2:58:05
practices the focus has been more on
2:58:05
practices the focus has been more on data analysis there
2:58:06
data analysis there
2:58:06
data analysis there the other type of data scientist that i
2:58:09
the other type of data scientist that i
2:58:09
the other type of data scientist that i encounter in a while is the is the
2:58:10
encounter in a while is the is the
2:58:10
encounter in a while is the is the person that's a b
2:58:11
person that's a b
2:58:11
person that's a b type a builder if you will the person
2:58:14
type a builder if you will the person
2:58:14
type a builder if you will the person that's actually coming from a software
2:58:16
that's actually coming from a software
2:58:16
that's actually coming from a software engineering background
2:58:17
engineering background
2:58:17
engineering background and thinking in terms of i need to test
2:58:19
and thinking in terms of i need to test
2:58:20
and thinking in terms of i need to test this i need to focus on a unit test
2:58:22
this i need to focus on a unit test
2:58:22
this i need to focus on a unit test first and then write
2:58:23
first and then write
2:58:23
first and then write my model we actually started sort of
2:58:26
my model we actually started sort of
2:58:26
my model we actually started sort of selecting towards csb type
2:58:27
selecting towards csb type
2:58:28
selecting towards csb type and teaching people all these software
2:58:30
and teaching people all these software
2:58:30
and teaching people all these software engineering
2:58:31
engineering
2:58:31
engineering skills because that turns out to be very
2:58:33
skills because that turns out to be very
2:58:33
skills because that turns out to be very important
2:58:34
important
2:58:34
important to not only get people that are writing
2:58:37
to not only get people that are writing
2:58:37
to not only get people that are writing need code
2:58:37
need code
2:58:38
need code because we we've seen some pretty
2:58:39
because we we've seen some pretty
2:58:40
because we we've seen some pretty interesting python notebooks
2:58:42
interesting python notebooks
2:58:42
interesting python notebooks producing models that's sort of a big
2:58:44
producing models that's sort of a big
2:58:44
producing models that's sort of a big new in production
2:58:45
new in production
2:58:46
new in production um but also getting people into the
2:58:49
um but also getting people into the
2:58:49
um but also getting people into the mindset of
2:58:50
mindset of
2:58:50
mindset of testing first writing a test first and
2:58:52
testing first writing a test first and
2:58:52
testing first writing a test first and then
2:58:53
then
2:58:53
then writing the code to actually generate
2:58:55
writing the code to actually generate
2:58:55
writing the code to actually generate this model instead of the other way
2:58:57
this model instead of the other way
2:58:57
this model instead of the other way around
2:58:57
around
2:58:58
around it's really been helpful so um
2:59:01
it's really been helpful so um
2:59:01
it's really been helpful so um and noelle have you seen this also in
2:59:03
and noelle have you seen this also in
2:59:03
and noelle have you seen this also in other companies actually that you've
2:59:05
other companies actually that you've
2:59:05
other companies actually that you've worked with
2:59:07
worked with
2:59:07
worked with yeah well it's definitely a tendency or
2:59:09
yeah well it's definitely a tendency or
2:59:09
yeah well it's definitely a tendency or a
2:59:10
a
2:59:10
a direction i'm encouraging organizations
2:59:13
direction i'm encouraging organizations
2:59:13
direction i'm encouraging organizations to go in
2:59:13
to go in
2:59:14
to go in mainly because i actually don't we have
2:59:16
mainly because i actually don't we have
2:59:16
mainly because i actually don't we have data scientists
2:59:17
data scientists
2:59:17
data scientists and a lot of them were sitting in an r d
2:59:20
and a lot of them were sitting in an r d
2:59:20
and a lot of them were sitting in an r d type team
2:59:21
type team
2:59:21
type team and so productizing an r d team is not
2:59:24
and so productizing an r d team is not
2:59:24
and so productizing an r d team is not necessarily
2:59:25
necessarily
2:59:25
necessarily the right thing to do for every
2:59:26
the right thing to do for every
2:59:26
the right thing to do for every organization oftentimes i like to see
2:59:29
organization oftentimes i like to see
2:59:29
organization oftentimes i like to see that r
2:59:29
that r
2:59:29
that r d team do their thing keep researching
2:59:32
d team do their thing keep researching
2:59:32
d team do their thing keep researching and then they deploy their model
2:59:34
and then they deploy their model
2:59:34
and then they deploy their model and it's handed to what i have now
2:59:36
and it's handed to what i have now
2:59:36
and it's handed to what i have now created
2:59:37
created
2:59:37
created ai engineers right those builders that
2:59:40
ai engineers right those builders that
2:59:40
ai engineers right those builders that that understand machine learning that
2:59:41
that understand machine learning that
2:59:41
that understand machine learning that are
2:59:42
are
2:59:42
are classically trained if you will but that
2:59:44
classically trained if you will but that
2:59:44
classically trained if you will but that are more software engineering focus
2:59:46
are more software engineering focus
2:59:46
are more software engineering focus they are the ones that build and use the
2:59:48
they are the ones that build and use the
2:59:48
they are the ones that build and use the code pipelines they're the ones who
2:59:50
code pipelines they're the ones who
2:59:50
code pipelines they're the ones who manage
2:59:50
manage
2:59:50
manage change in production and so i do have
2:59:53
change in production and so i do have
2:59:53
change in production and so i do have that same separation and i've seen it
2:59:54
that same separation and i've seen it
2:59:54
that same separation and i've seen it work well
2:59:55
work well
2:59:55
work well but i also know very large companies
2:59:57
but i also know very large companies
2:59:57
but i also know very large companies like microsoft and amazon
2:59:59
like microsoft and amazon
2:59:59
like microsoft and amazon have taken their r d teams and just
3:00:00
have taken their r d teams and just
3:00:00
have taken their r d teams and just picked them up and dropped them into
3:00:02
picked them up and dropped them into
3:00:02
picked them up and dropped them into productized organizations
3:00:04
productized organizations
3:00:04
productized organizations and that seems to be working okay too so
3:00:07
and that seems to be working okay too so
3:00:07
and that seems to be working okay too so um
3:00:07
um
3:00:08
um yeah so i i think there's probably
3:00:09
yeah so i i think there's probably
3:00:09
yeah so i i think there's probably benefits to both but
3:00:11
benefits to both but
3:00:11
benefits to both but definitely understanding that there's so
3:00:12
definitely understanding that there's so
3:00:12
definitely understanding that there's so much that we can learn from both sides
3:00:14
much that we can learn from both sides
3:00:14
much that we can learn from both sides the analytical mind
3:00:16
the analytical mind
3:00:16
the analytical mind and i actually at hacker u i have a full
3:00:19
and i actually at hacker u i have a full
3:00:19
and i actually at hacker u i have a full stack
3:00:19
stack
3:00:19
stack development team and a data science team
3:00:21
development team and a data science team
3:00:21
development team and a data science team and they think
3:00:22
and they think
3:00:22
and they think completely differently but that
3:00:24
completely differently but that
3:00:24
completely differently but that diversity of thought
3:00:26
diversity of thought
3:00:26
diversity of thought creates much better outcomes and i think
3:00:28
creates much better outcomes and i think
3:00:28
creates much better outcomes and i think that that's probably
3:00:29
that that's probably
3:00:29
that that's probably the moral of the story from my
3:00:31
the moral of the story from my
3:00:31
the moral of the story from my perspective yeah
3:00:32
perspective yeah
3:00:32
perspective yeah that makes sense so there's not actually
3:00:35
that makes sense so there's not actually
3:00:35
that makes sense so there's not actually one
3:00:36
one
3:00:36
one shape of data scientists or one type of
3:00:39
shape of data scientists or one type of
3:00:39
shape of data scientists or one type of team that's that's going to get you
3:00:41
team that's that's going to get you
3:00:41
team that's that's going to get you across the finish line
3:00:42
across the finish line
3:00:42
across the finish line there's um it's important to keep this
3:00:45
there's um it's important to keep this
3:00:45
there's um it's important to keep this mindset
3:00:46
mindset
3:00:46
mindset regardless of what you have in your
3:00:48
regardless of what you have in your
3:00:48
regardless of what you have in your company
3:00:51
we we need to make sure that people
3:00:53
we we need to make sure that people
3:00:53
we we need to make sure that people understand that it starts with the
3:00:55
understand that it starts with the
3:00:55
understand that it starts with the asking the right questions to your model
3:00:59
yeah okay that's ivory
3:01:03
cool yeah it should always start with a
3:01:05
cool yeah it should always start with a
3:01:05
cool yeah it should always start with a with a sharp question
3:01:06
with a sharp question
3:01:06
with a sharp question what is it no one would ever like be
3:01:09
what is it no one would ever like be
3:01:09
what is it no one would ever like be like oh i need to write a function today
3:01:11
like oh i need to write a function today
3:01:11
like oh i need to write a function today uh just like people are like oh i need
3:01:13
uh just like people are like oh i need
3:01:13
uh just like people are like oh i need to make some ai today no that's
3:01:15
to make some ai today no that's
3:01:15
to make some ai today no that's it's it's the same thing you don't write
3:01:17
it's it's the same thing you don't write
3:01:17
it's it's the same thing you don't write functions for no reason
3:01:18
functions for no reason
3:01:18
functions for no reason and a lot of people right now are like
3:01:20
and a lot of people right now are like
3:01:20
and a lot of people right now are like running around with a hammer which is
3:01:22
running around with a hammer which is
3:01:22
running around with a hammer which is the ml machine learning handle hammer
3:01:24
the ml machine learning handle hammer
3:01:24
the ml machine learning handle hammer and they're like what should i
3:01:25
and they're like what should i
3:01:25
and they're like what should i hit with it and they're trying to like
3:01:28
hit with it and they're trying to like
3:01:28
hit with it and they're trying to like hit screws in with a hammer
3:01:29
hit screws in with a hammer
3:01:29
hit screws in with a hammer sure you'll get it to go in but you're
3:01:31
sure you'll get it to go in but you're
3:01:31
sure you'll get it to go in but you're going to break some stuff right and
3:01:32
going to break some stuff right and
3:01:32
going to break some stuff right and maybe
3:01:33
maybe
3:01:33
maybe machine learning isn't the right
3:01:35
machine learning isn't the right
3:01:35
machine learning isn't the right approach
3:01:36
approach
3:01:36
approach uh but you should always start with a
3:01:37
uh but you should always start with a
3:01:37
uh but you should always start with a sharp question right is it is it
3:01:39
sharp question right is it is it
3:01:39
sharp question right is it is it are we trying to guess between are we
3:01:40
are we trying to guess between are we
3:01:40
are we trying to guess between are we trying to predict between classes
3:01:42
trying to predict between classes
3:01:42
trying to predict between classes are we trying to get a you know a
3:01:45
are we trying to get a you know a
3:01:45
are we trying to get a you know a continuous value
3:01:47
continuous value
3:01:47
continuous value do we even have a value that we want to
3:01:49
do we even have a value that we want to
3:01:49
do we even have a value that we want to come out with and
3:01:50
come out with and
3:01:50
come out with and it should start with a business question
3:01:53
it should start with a business question
3:01:53
it should start with a business question and the reality matter is the people
3:01:54
and the reality matter is the people
3:01:54
and the reality matter is the people that are successfully using machine
3:01:56
that are successfully using machine
3:01:56
that are successfully using machine learning
3:01:57
learning
3:01:57
learning don't talk about it that much because if
3:01:59
don't talk about it that much because if
3:01:59
don't talk about it that much because if they were to tell me
3:02:00
they were to tell me
3:02:00
they were to tell me hey seth i made a machine learning model
3:02:02
hey seth i made a machine learning model
3:02:02
hey seth i made a machine learning model that did foo i would
3:02:03
that did foo i would
3:02:04
that did foo i would almost immediately know how they built
3:02:05
almost immediately know how they built
3:02:05
almost immediately know how they built it almost
3:02:07
it almost
3:02:07
it almost like pretty close right and so the
3:02:09
like pretty close right and so the
3:02:09
like pretty close right and so the people that are using it well
3:02:11
people that are using it well
3:02:11
people that are using it well have an understanding there's a business
3:02:12
have an understanding there's a business
3:02:12
have an understanding there's a business need for it have an understanding that
3:02:14
need for it have an understanding that
3:02:14
need for it have an understanding that there's responsibility behind using it
3:02:16
there's responsibility behind using it
3:02:16
there's responsibility behind using it and they probably have some engineering
3:02:18
and they probably have some engineering
3:02:18
and they probably have some engineering practices behind it
3:02:20
practices behind it
3:02:20
practices behind it um so if you want to get to that point
3:02:23
um so if you want to get to that point
3:02:23
um so if you want to get to that point it's all
3:02:24
it's all
3:02:24
it's all programmer stuff everyone knows all
3:02:26
programmer stuff everyone knows all
3:02:26
programmer stuff everyone knows all programmers know how to do
3:02:27
programmers know how to do
3:02:27
programmers know how to do this this kind of thing huh that's
3:02:31
this this kind of thing huh that's
3:02:31
this this kind of thing huh that's pretty cool
3:02:31
pretty cool
3:02:32
pretty cool i mean yeah and it's kind of nice
3:02:33
i mean yeah and it's kind of nice
3:02:33
i mean yeah and it's kind of nice because it allows us to open up
3:02:35
because it allows us to open up
3:02:35
because it allows us to open up opportunities for reskilling our staff
3:02:37
opportunities for reskilling our staff
3:02:37
opportunities for reskilling our staff right like i know some of us in the data
3:02:40
right like i know some of us in the data
3:02:40
right like i know some of us in the data science world the longer we've been
3:02:42
science world the longer we've been
3:02:42
science world the longer we've been in it the more pristine that environment
3:02:45
in it the more pristine that environment
3:02:45
in it the more pristine that environment and echo chambery that environment is
3:02:47
and echo chambery that environment is
3:02:47
and echo chambery that environment is but the reality is now i mean anthony
3:02:49
but the reality is now i mean anthony
3:02:49
but the reality is now i mean anthony was a great example like the reality is
3:02:51
was a great example like the reality is
3:02:51
was a great example like the reality is now people who have a problem
3:02:53
now people who have a problem
3:02:53
now people who have a problem if they're technically inclined we have
3:02:56
if they're technically inclined we have
3:02:56
if they're technically inclined we have technology that allows them
3:02:57
technology that allows them
3:02:58
technology that allows them to solve that problem or at least build
3:02:59
to solve that problem or at least build
3:02:59
to solve that problem or at least build the poc so that they can articulate that
3:03:01
the poc so that they can articulate that
3:03:02
the poc so that they can articulate that problem
3:03:02
problem
3:03:02
problem uh in the most accurate way to a modeler
3:03:05
uh in the most accurate way to a modeler
3:03:05
uh in the most accurate way to a modeler and so i feel like really
3:03:06
and so i feel like really
3:03:06
and so i feel like really closing that gap between the business
3:03:08
closing that gap between the business
3:03:08
closing that gap between the business user and
3:03:09
user and
3:03:09
user and the technologist that's going to solve
3:03:11
the technologist that's going to solve
3:03:11
the technologist that's going to solve it is something that we're seeing
3:03:13
it is something that we're seeing
3:03:13
it is something that we're seeing at least in my career over the last 20
3:03:15
at least in my career over the last 20
3:03:15
at least in my career over the last 20 years i've never seen it be
3:03:17
years i've never seen it be
3:03:17
years i've never seen it be as possible and successful as i do today
3:03:20
as possible and successful as i do today
3:03:20
as possible and successful as i do today now that we've got these kind of
3:03:21
now that we've got these kind of
3:03:21
now that we've got these kind of democratized models and ability for
3:03:23
democratized models and ability for
3:03:23
democratized models and ability for people to use a web browser
3:03:24
people to use a web browser
3:03:24
people to use a web browser to train a model they want to train the
3:03:26
to train a model they want to train the
3:03:26
to train a model they want to train the model but they can at least give me the
3:03:28
model but they can at least give me the
3:03:28
model but they can at least give me the right idea
3:03:29
right idea
3:03:29
right idea um as my team actually builds it so i
3:03:32
um as my team actually builds it so i
3:03:32
um as my team actually builds it so i think that's unique
3:03:33
think that's unique
3:03:33
think that's unique and something we can be grateful for
3:03:35
and something we can be grateful for
3:03:35
and something we can be grateful for yeah certainly yeah
3:03:37
yeah certainly yeah
3:03:37
yeah certainly yeah so actually what you're saying is don't
3:03:39
so actually what you're saying is don't
3:03:39
so actually what you're saying is don't worry so much about the fact that
3:03:41
worry so much about the fact that
3:03:41
worry so much about the fact that a data scientist has formula formal
3:03:44
a data scientist has formula formal
3:03:44
a data scientist has formula formal education
3:03:45
education
3:03:45
education in statistics and all the math that you
3:03:47
in statistics and all the math that you
3:03:47
in statistics and all the math that you need to build a model from scratch
3:03:49
need to build a model from scratch
3:03:50
need to build a model from scratch worry about and worry about the fact
3:03:51
worry about and worry about the fact
3:03:51
worry about and worry about the fact that you know as a software engineer
3:03:53
that you know as a software engineer
3:03:53
that you know as a software engineer maybe about
3:03:54
maybe about
3:03:54
maybe about the business that your customer is in
3:03:57
the business that your customer is in
3:03:57
the business that your customer is in and then start
3:03:58
and then start
3:03:58
and then start working with the democratized tools that
3:04:01
working with the democratized tools that
3:04:01
working with the democratized tools that we have such as cognitive services
3:04:03
we have such as cognitive services
3:04:03
we have such as cognitive services and at least get started that's really
3:04:05
and at least get started that's really
3:04:05
and at least get started that's really cool it breaks down the gates to
3:04:07
cool it breaks down the gates to
3:04:07
cool it breaks down the gates to this amazing technology that we as a
3:04:09
this amazing technology that we as a
3:04:09
this amazing technology that we as a community have been talking about for
3:04:11
community have been talking about for
3:04:11
community have been talking about for years now
3:04:12
years now
3:04:12
years now um yeah that that's
3:04:15
um yeah that that's
3:04:15
um yeah that that's really cool to hear that from you so
3:04:17
really cool to hear that from you so
3:04:17
really cool to hear that from you so yeah i mean not all of us can talk like
3:04:19
yeah i mean not all of us can talk like
3:04:19
yeah i mean not all of us can talk like seth
3:04:21
seth
3:04:21
seth he's pretty fast with all of that
3:04:24
he's pretty fast with all of that
3:04:24
he's pretty fast with all of that on people like seth to solve all of the
3:04:26
on people like seth to solve all of the
3:04:26
on people like seth to solve all of the problems like
3:04:27
problems like
3:04:28
problems like we wouldn't get there we can't get there
3:04:30
we wouldn't get there we can't get there
3:04:30
we wouldn't get there we can't get there so we have to
3:04:31
so we have to
3:04:31
so we have to figure out a way to get at least started
3:04:33
figure out a way to get at least started
3:04:33
figure out a way to get at least started like just get your problem
3:04:35
like just get your problem
3:04:35
like just get your problem get it like a modeled in a democratized
3:04:38
get it like a modeled in a democratized
3:04:38
get it like a modeled in a democratized model so that someone like
3:04:39
model so that someone like
3:04:40
model so that someone like you know an engineering team a model
3:04:42
you know an engineering team a model
3:04:42
you know an engineering team a model engineering team can look at and go
3:04:44
engineering team can look at and go
3:04:44
engineering team can look at and go let's see what you're trying to do there
3:04:45
let's see what you're trying to do there
3:04:45
let's see what you're trying to do there okay i can work with this
3:04:47
okay i can work with this
3:04:47
okay i can work with this as opposed to right now you know one
3:04:50
as opposed to right now you know one
3:04:50
as opposed to right now you know one side says something and the other side
3:04:52
side says something and the other side
3:04:52
side says something and the other side here's something totally different
3:04:54
here's something totally different
3:04:54
here's something totally different but now we've got tools and technologies
3:04:55
but now we've got tools and technologies
3:04:56
but now we've got tools and technologies that really bridge that gap and i'm tr
3:04:57
that really bridge that gap and i'm tr
3:04:57
that really bridge that gap and i'm tr you know that's what i do now is i go
3:04:59
you know that's what i do now is i go
3:04:59
you know that's what i do now is i go talk to as many companies who will hear
3:05:00
talk to as many companies who will hear
3:05:00
talk to as many companies who will hear me
3:05:01
me
3:05:01
me to tell them like you don't have to wait
3:05:03
to tell them like you don't have to wait
3:05:03
to tell them like you don't have to wait for your data scientists to show up like
3:05:05
for your data scientists to show up like
3:05:05
for your data scientists to show up like you can start right now
3:05:06
you can start right now
3:05:06
you can start right now you can do this right now with your web
3:05:08
you can do this right now with your web
3:05:08
you can do this right now with your web developers and software engineers
3:05:10
developers and software engineers
3:05:10
developers and software engineers they can do quite a bit to get you going
3:05:12
they can do quite a bit to get you going
3:05:12
they can do quite a bit to get you going um and then
3:05:13
um and then
3:05:13
um and then you know as you invest and as you find
3:05:15
you know as you invest and as you find
3:05:15
you know as you invest and as you find the data scientists to help you finish
3:05:17
the data scientists to help you finish
3:05:17
the data scientists to help you finish the deal
3:05:17
the deal
3:05:18
the deal the job so i've got and i am not a
3:05:21
the job so i've got and i am not a
3:05:21
the job so i've got and i am not a research
3:05:21
research
3:05:22
research caliber data guy fyi i'm not
3:05:25
caliber data guy fyi i'm not
3:05:25
caliber data guy fyi i'm not i just like the math right and so i i
3:05:28
i just like the math right and so i i
3:05:28
i just like the math right and so i i like understanding how
3:05:29
like understanding how
3:05:29
like understanding how liking math is enough
3:05:32
liking math is enough
3:05:32
liking math is enough i just like how it works like internally
3:05:34
i just like how it works like internally
3:05:34
i just like how it works like internally like once you know how it works it's
3:05:35
like once you know how it works it's
3:05:35
like once you know how it works it's quite impressive
3:05:37
quite impressive
3:05:37
quite impressive that it's actually doing anything
3:05:38
that it's actually doing anything
3:05:38
that it's actually doing anything reasonable uh which
3:05:40
reasonable uh which
3:05:40
reasonable uh which is to be super surprising and so i
3:05:42
is to be super surprising and so i
3:05:42
is to be super surprising and so i wouldn't consider myself to be
3:05:44
wouldn't consider myself to be
3:05:44
wouldn't consider myself to be on the avant-garde i actually
3:05:45
on the avant-garde i actually
3:05:45
on the avant-garde i actually interviewed people that are like i've
3:05:47
interviewed people that are like i've
3:05:47
interviewed people that are like i've interviewed touring award winners
3:05:49
interviewed touring award winners
3:05:49
interviewed touring award winners that have done deep learning for a
3:05:51
that have done deep learning for a
3:05:51
that have done deep learning for a living and research and
3:05:52
living and research and
3:05:52
living and research and they too say stuff like you know it's
3:05:55
they too say stuff like you know it's
3:05:55
they too say stuff like you know it's still kind of squishy like for example
3:05:56
still kind of squishy like for example
3:05:56
still kind of squishy like for example with earlier machine learning models you
3:05:58
with earlier machine learning models you
3:05:58
with earlier machine learning models you could actually use uh
3:06:00
could actually use uh
3:06:00
could actually use uh standard mathematical approaches to
3:06:02
standard mathematical approaches to
3:06:02
standard mathematical approaches to understand how good they are
3:06:04
understand how good they are
3:06:04
understand how good they are you can't do that with deep learning at
3:06:06
you can't do that with deep learning at
3:06:06
you can't do that with deep learning at all right now
3:06:07
all right now
3:06:07
all right now uh there's no mathematical framework for
3:06:09
uh there's no mathematical framework for
3:06:09
uh there's no mathematical framework for for them
3:06:10
for them
3:06:10
for them for us to predict those kinds of things
3:06:12
for us to predict those kinds of things
3:06:12
for us to predict those kinds of things which is a telling
3:06:15
which is a telling
3:06:15
which is a telling yeah yeah you i think the whole
3:06:19
yeah yeah you i think the whole
3:06:19
yeah yeah you i think the whole approach between the neat people and
3:06:21
approach between the neat people and
3:06:21
approach between the neat people and scruffy people i i remember that from
3:06:23
scruffy people i i remember that from
3:06:23
scruffy people i i remember that from the first episode we had a conversation
3:06:24
the first episode we had a conversation
3:06:24
the first episode we had a conversation with richard about that
3:06:26
with richard about that
3:06:26
with richard about that um yeah right now we're in in a episode
3:06:29
um yeah right now we're in in a episode
3:06:29
um yeah right now we're in in a episode where we can see
3:06:30
where we can see
3:06:30
where we can see that the scruffy people have the the
3:06:32
that the scruffy people have the the
3:06:32
that the scruffy people have the the upper hand
3:06:34
upper hand
3:06:34
upper hand um and and it feels to me that uh things
3:06:36
um and and it feels to me that uh things
3:06:36
um and and it feels to me that uh things like explainers and interpreters are
3:06:38
like explainers and interpreters are
3:06:38
like explainers and interpreters are more important than ever
3:06:39
more important than ever
3:06:40
more important than ever actually to make sense of all the stuff
3:06:41
actually to make sense of all the stuff
3:06:42
actually to make sense of all the stuff that that they are doing
3:06:43
that that they are doing
3:06:43
that that they are doing um so thank you very much both of you
3:06:47
um so thank you very much both of you
3:06:47
um so thank you very much both of you we're we're we're full out of time this
3:06:49
we're we're we're full out of time this
3:06:49
we're we're we're full out of time this is going so fast so if
3:06:51
is going so fast so if
3:06:51
is going so fast so if if people want to continue the
3:06:52
if people want to continue the
3:06:52
if people want to continue the conversation please do so in a live chat
3:06:56
conversation please do so in a live chat
3:06:56
conversation please do so in a live chat it will be active after the show as well
3:06:58
it will be active after the show as well
3:06:58
it will be active after the show as well um
3:06:59
um
3:06:59
um let me quickly look up because we've got
3:07:01
let me quickly look up because we've got
3:07:01
let me quickly look up because we've got a winner for
3:07:03
a winner for
3:07:03
a winner for the uh
3:07:06
for our prize if my computer wants to
3:07:09
for our prize if my computer wants to
3:07:09
for our prize if my computer wants to cooperate
3:07:15
so um vladimir you're you're the lucky
3:07:19
so um vladimir you're you're the lucky
3:07:19
so um vladimir you're you're the lucky person today
3:07:20
person today
3:07:20
person today uh you won the 50 amazon gift card for
3:07:22
uh you won the 50 amazon gift card for
3:07:22
uh you won the 50 amazon gift card for making sure that you that you get that
3:07:25
making sure that you that you get that
3:07:25
making sure that you that you get that um and alicia do you have any final
3:07:28
um and alicia do you have any final
3:07:28
um and alicia do you have any final thoughts we're at the end of the episode
3:07:30
thoughts we're at the end of the episode
3:07:30
thoughts we're at the end of the episode so this has been a fantastic set of
3:07:33
so this has been a fantastic set of
3:07:33
so this has been a fantastic set of sessions and i thank you to all of our
3:07:36
sessions and i thank you to all of our
3:07:36
sessions and i thank you to all of our speakers
3:07:37
speakers
3:07:37
speakers i i keep i'm amazed every time
3:07:40
i i keep i'm amazed every time
3:07:40
i i keep i'm amazed every time the global ai community puts on an event
3:07:43
the global ai community puts on an event
3:07:43
the global ai community puts on an event and the quality of speakers and the
3:07:45
and the quality of speakers and the
3:07:45
and the quality of speakers and the passion
3:07:46
passion
3:07:46
passion and i'm so grateful to be part of the
3:07:48
and i'm so grateful to be part of the
3:07:48
and i'm so grateful to be part of the microsoft community
3:07:50
microsoft community
3:07:50
microsoft community and i'm definitely looking forward to
3:07:53
and i'm definitely looking forward to
3:07:53
and i'm definitely looking forward to february with global ai boot camps
3:07:57
february with global ai boot camps
3:07:57
february with global ai boot camps so thank you for for including us
3:08:00
so thank you for for including us
3:08:00
so thank you for for including us yeah yeah you're welcome and
3:08:04
yeah yeah you're welcome and
3:08:04
yeah yeah you're welcome and i do have to thank all the speakers
3:08:07
i do have to thank all the speakers
3:08:07
i do have to thank all the speakers everyone
3:08:07
everyone
3:08:08
everyone who's joined us and spent time talking
3:08:10
who's joined us and spent time talking
3:08:10
who's joined us and spent time talking to us
3:08:11
to us
3:08:11
to us coming up with interview questions with
3:08:13
coming up with interview questions with
3:08:13
coming up with interview questions with ideas for sessions and checking out all
3:08:15
ideas for sessions and checking out all
3:08:15
ideas for sessions and checking out all the technical stuff
3:08:16
the technical stuff
3:08:16
the technical stuff um i have to say nobody's seen hank in
3:08:19
um i have to say nobody's seen hank in
3:08:19
um i have to say nobody's seen hank in the episodes
3:08:19
the episodes
3:08:20
the episodes but hank is behind the scenes he's right
3:08:21
but hank is behind the scenes he's right
3:08:21
but hank is behind the scenes he's right here in the studio he's doing all the
3:08:23
here in the studio he's doing all the
3:08:23
here in the studio he's doing all the technical stuff for us
3:08:25
technical stuff for us
3:08:25
technical stuff for us pushing buttons making sure that the
3:08:26
pushing buttons making sure that the
3:08:26
pushing buttons making sure that the sound is working uh failing video
3:08:28
sound is working uh failing video
3:08:28
sound is working uh failing video connections just like we had tonight
3:08:30
connections just like we had tonight
3:08:30
connections just like we had tonight with seth
3:08:32
with seth
3:08:32
with seth that's amazing too that's really
3:08:34
that's amazing too that's really
3:08:34
that's amazing too that's really important stuff that that needs to be
3:08:35
important stuff that that needs to be
3:08:36
important stuff that that needs to be done and
3:08:36
done and
3:08:36
done and uh i know how many hours he spent on on
3:08:39
uh i know how many hours he spent on on
3:08:39
uh i know how many hours he spent on on getting this
3:08:40
getting this
3:08:40
getting this all to work for us um and and finally i
3:08:44
all to work for us um and and finally i
3:08:44
all to work for us um and and finally i i'd like to thank our sponsors as well
3:08:46
i'd like to thank our sponsors as well
3:08:46
i'd like to thank our sponsors as well uh microsoft have been very grateful
3:08:50
to give us the money to make this all
3:08:53
to give us the money to make this all
3:08:53
to give us the money to make this all possible they've helped us in many many
3:08:55
possible they've helped us in many many
3:08:55
possible they've helped us in many many ways
3:08:57
ways
3:08:57
ways speakers also contacts in the field
3:09:01
speakers also contacts in the field
3:09:01
speakers also contacts in the field uh my own employer your employer alicia
3:09:04
uh my own employer your employer alicia
3:09:04
uh my own employer your employer alicia for giving us time to do this actually
3:09:06
for giving us time to do this actually
3:09:06
for giving us time to do this actually during during our work days
3:09:08
during during our work days
3:09:08
during during our work days uh that's not possible for everyone and
3:09:10
uh that's not possible for everyone and
3:09:10
uh that's not possible for everyone and i'm especially grateful that we can do
3:09:12
i'm especially grateful that we can do
3:09:12
i'm especially grateful that we can do that
3:09:13
that
3:09:13
that to offer the community the best possible
3:09:15
to offer the community the best possible
3:09:15
to offer the community the best possible content um
3:09:17
content um
3:09:17
content um at least i think we've we've uh spent a
3:09:20
at least i think we've we've uh spent a
3:09:20
at least i think we've we've uh spent a lot of time and and effort to do this
3:09:22
lot of time and and effort to do this
3:09:22
lot of time and and effort to do this and and i hope people like it
3:09:24
and and i hope people like it
3:09:24
and and i hope people like it um so yeah a huge thank you for everyone
3:09:28
um so yeah a huge thank you for everyone
3:09:28
um so yeah a huge thank you for everyone for everyone watching we've
3:09:29
for everyone watching we've
3:09:29
for everyone watching we've recent had so many reactions i mean it's
3:09:32
recent had so many reactions i mean it's
3:09:32
recent had so many reactions i mean it's unbelievable how many people have asked
3:09:34
unbelievable how many people have asked
3:09:34
unbelievable how many people have asked us questions send us comments afterwards
3:09:36
us questions send us comments afterwards
3:09:36
us questions send us comments afterwards uh thanking us
3:09:38
uh thanking us
3:09:38
uh thanking us um so i'm happy and with that yeah this
3:09:41
um so i'm happy and with that yeah this
3:09:41
um so i'm happy and with that yeah this is the
3:09:42
is the
3:09:42
is the last episode of the october sessions so
3:09:45
last episode of the october sessions so
3:09:45
last episode of the october sessions so um
3:09:45
um
3:09:45
um we'll be gone for a little bit for
3:09:47
we'll be gone for a little bit for
3:09:48
we'll be gone for a little bit for christmas time possibly
3:09:49
christmas time possibly
3:09:49
christmas time possibly but we'll be back so we have globally
3:09:52
but we'll be back so we have globally
3:09:52
but we'll be back so we have globally boot camps
3:09:53
boot camps
3:09:53
boot camps in in january february people can
3:09:57
in in january february people can
3:09:57
in in january february people can join those those are locally organized
3:09:59
join those those are locally organized
3:09:59
join those those are locally organized by a lot of people across the globe
3:10:01
by a lot of people across the globe
3:10:01
by a lot of people across the globe uh i know there's a hundred plus
3:10:03
uh i know there's a hundred plus
3:10:03
uh i know there's a hundred plus locations so that that's a lot of
3:10:04
locations so that that's a lot of
3:10:04
locations so that that's a lot of places where you can go to learn more
3:10:06
places where you can go to learn more
3:10:06
places where you can go to learn more about ai
3:10:07
about ai
3:10:08
about ai and i'm especially looking forward to
3:10:09
and i'm especially looking forward to
3:10:09
and i'm especially looking forward to people trying things out for example
3:10:12
people trying things out for example
3:10:12
people trying things out for example building your own custom computer vision
3:10:14
building your own custom computer vision
3:10:14
building your own custom computer vision model maybe trying out pytorch please
3:10:16
model maybe trying out pytorch please
3:10:16
model maybe trying out pytorch please let us know on twitter
3:10:18
let us know on twitter
3:10:18
let us know on twitter to our usual channels send us a message
3:10:21
to our usual channels send us a message
3:10:21
to our usual channels send us a message and show us your stuff
3:10:22
and show us your stuff
3:10:22
and show us your stuff that you're working on i'm certainly
3:10:24
that you're working on i'm certainly
3:10:24
that you're working on i'm certainly curious and if
3:10:26
curious and if
3:10:26
curious and if if anybody has anything interesting to
3:10:28
if anybody has anything interesting to
3:10:28
if anybody has anything interesting to to tell
3:10:29
to tell
3:10:29
to tell tell us i'm sure sammy who's been here
3:10:32
tell us i'm sure sammy who's been here
3:10:32
tell us i'm sure sammy who's been here on the episode as well
3:10:33
on the episode as well
3:10:34
on the episode as well is happy to have them on the global ai
3:10:35
is happy to have them on the global ai
3:10:35
is happy to have them on the global ai talks which is a weekly show that we run
3:10:37
talks which is a weekly show that we run
3:10:37
talks which is a weekly show that we run on thursday
3:10:38
on thursday
3:10:38
on thursday also which will continue next week and
3:10:41
also which will continue next week and
3:10:41
also which will continue next week and pick up all the stories that we've been
3:10:42
pick up all the stories that we've been
3:10:42
pick up all the stories that we've been talking about
3:10:43
talking about
3:10:43
talking about with other speakers uh in the field so
3:10:46
with other speakers uh in the field so
3:10:46
with other speakers uh in the field so yeah thank you very much alicia for
3:10:48
yeah thank you very much alicia for
3:10:48
yeah thank you very much alicia for joining us how's the puppy doing
3:10:52
joining us how's the puppy doing
3:10:52
joining us how's the puppy doing i think he's napping right now
3:10:56
thankfully
3:10:57
thankfully
3:10:57
thankfully [Laughter]
3:10:59
[Laughter]
3:10:59
[Laughter] can i still adopt him absolutely
3:11:02
can i still adopt him absolutely
3:11:02
can i still adopt him absolutely um you know what i will stick him in my
3:11:05
um you know what i will stick him in my
3:11:05
um you know what i will stick him in my suitcase
3:11:06
suitcase
3:11:06
suitcase when i come visit you oh yeah
3:11:09
when i come visit you oh yeah
3:11:09
when i come visit you oh yeah that would be awesome well thank you
3:11:11
that would be awesome well thank you
3:11:11
that would be awesome well thank you very much everyone have a nice
3:11:13
very much everyone have a nice
3:11:14
very much everyone have a nice uh afternoon uh morning evening
3:11:17
uh afternoon uh morning evening
3:11:17
uh afternoon uh morning evening uh wherever you are and stay safe and i
3:11:20
uh wherever you are and stay safe and i
3:11:20
uh wherever you are and stay safe and i will see you soon
3:11:22
will see you soon
3:11:22
will see you soon yes please everyone be safe see ya
3:11:38
[Music]
3:11:39
[Music]
3:11:39
[Music] you