A bit of AI- S02 - Ep 9

Thank you

Hi everyone, and welcome to the ninth episode of season two of the A Bit of AI show

This episode, we will talk with Ben about his life in AI

Hi everyone, and welcome to the ninth episode of season two of the A Bit of AI show

My name is Hank Boman. I'm a white male with brown hair, wearing glasses, and today I'm

wearing a black hoodie with the Azure Life logo on it. Very good description, Hank. And hi, my name is Amy Boyd. I am a white female with blonde

straight hair, and today I'm wearing a dark gray jumper. Mine's a little plainer, but

Hank, we didn't follow our usual rules. We're meant to be wearing bright clothing

and both of us turn up, but it is very much winter here

So I do apologize. Yeah, it's the time you just want to blend

into the background. So hi everyone. And once again, welcome to the A Bit of the AI Show

This show is all about the story from the people behind the AI systems

Because in our job, Amy and me, when we travel the world, meet so many people around the globe

And we realized that there are so many different skilled people involved in creating an AI solution

So from pre-sales to consultancy, to the people creating the deep neural networks itself

and running them in production. So in this show, we invite people from all over the world

that are professionals in the AI space to just have a chat about what they actually do

during the week and how they reach this point in their career

So this is the ninth episode of season two, and we couldn't be more excited

to start talking to our guests and learn more about what they do

And as always, everything is available on our website, abitofai.show. And you know what it's like

I'm all about the logistics. So after the show, you can join us

in the Abitofai experience. That's where you get to meet myself, Henk

and our wonderful guests straight after the show if you go to aka.ms slash a bit of AI dash cafe or if you're on

our website there's a button at the top of the screen. Perfect. So let's get started and invite Ben to the screen

Hi, how are you? Hi Ben, great to see you. Thank you so much for joining us. How are you today

great great how are you very very good well let's get kicked off because we

know how quick 30 minutes goes and we don't have that much time but I know you

have a lot to tell us so first things first Ben tell us who you are and what

you do in the space of AI okay so I am Ben Fishman and I am an algorithm

I'm a developer and data science and AI researcher. I work mainly with unstructured data

mainly with computer vision, audio signals, et cetera. And today I'm working for Microsoft in the Surface group

and I'm leading data science and algorithmic team. Amazing. Well, thank you for everything you do

I actually have a surface that I'm running on right now. So that's always a win

To know that there's a data science team always improving the behind the scenes is so fantastic

So you broke down sort of the types of data you work with there

So that was interesting because audio is quite a specific type of data. Same with video

And all jobs in our industry look completely different daily. But I was wondering, can you describe to us

What does your average day look like? Because you're both a data scientist, but you're also a manager

That's right. And actually it is quite different to be a data scientist and IC or a manager

And you look a little differently on the day to day. Your day to day looks differently

And I think that it starts at my position today, starting from the single task, which I'm not going to implement and research, but I need to follow and supervise in a way

So thinking about what will be the goal, how long it will take, what will be the different risk on it

and try to think of the general pipeline and architecture that we are going to use, not only in deep learning

but generally what will be the algorithm i can say that one of the main challenges is to take this

research field into production this is a big challenge today and therefore you know to be in a

production and a industry environment which i used to but you work with pms and software engineers

and other people with different skill sets. And you need to take this research field into specific, you know, time slots

and you have like deadlines, et cetera. And this is the main challenge of how to get it to work

And then after the specific task, I need to oversee of the team

the task and the people in the team. I have a lot of mentoring. It's not task, but you know

to go with the person, with the team member, to try to see what is the struggles, what are the

challenges and get into the details. Sometimes it's not, you know, only how long it's going to take

just predict you know the times etc You need to really get into the details of many many tasks because you know in the team there are many tasks and try not only always to solve it but sometimes you know just to guide or to ask the right question or to try to get back to the beginning and to think what are we looking for

And this is kind of the work that I'm doing in mentoring

Another very important thing is looking forward and to try to build the team, both technically

and from the HR perspective, because sometimes our tasks are very long-term

And if I want that, or if the team need to do a task in one or two years, maybe you need

to try to understand and get into a new fields, a new domain expertise

Maybe you need to bring these people into your team. And this is from the more HRE and domain expertise knowledge, but also we have all the technical

side of the infrastructure and build a team that know to work together, that have a common

research methodologies and to know to use one each, you know, ideas, code algorithms

to base on one each other. And this is part of my job as a team manager to try to create these collaborations and this

togetherness, filling in way of work. I really like that. That's a good perspective in some sense

I'm not sure it's one that we've had described quite like that way

You've kind of said I have day to day sort of more technical tasks and then we have to be collective as a team in solving those tasks

And then I always have to think about the long term. What's coming down the line

what do we want to achieve in the future and skill up my team or maybe I need to go find someone that's going to help us as a team get in the right place

Goodness me. There's a lot of spinning plates, as people say, on your plate there, Van

Definitely. Absolutely. Sometimes I get up in the middle of the night and think, oh, this algorithm, we need to

solve it like this or this type, we need to solve it

So yeah, there is lots on my mind. Amazing. Yeah. So you also used IC and maybe our viewers don't know what an IC is

So that is individual contributor via as being a manager. I think that's what you meant

Perfect. Yeah, perfect. So what I'm wondering is like you went from individual contributor to manager

but you also went to school, you studied something. So I want to hear about your journey

How did you get to the point where you're actually a manager and managing data science teams

Yeah, okay. So a long time ago, it wasn't that long time ago, but I started actually

as a biomedical engineer. I started to learn it because at one point, I really liked to

be an engineer, to build something new and I really liked the biomedical fields as well

But on the, you know, when I started, it was the bachelor, I went more and more to the electrical engineering part and the signal processing

Back then I called it signal processing and tried to do algorithms

Back then I already liked, you know, the image and image processing, computer vision and the audio part

and I call it algorithms and signal processing, but that was like the first steps, I think

Then I started, you know, the journey in the industry and I started in a company called Mobileye

Now it's part of Intel. They developed developing the autonomous vehicles. So they're well known

And I was there an algorithm development. there was a very big department of algorithms

And then I started to get to know and to learn how to get these algorithms

and started with ML as well into the real life. Okay, how to solve a real life problem

So I started from more classical solution and ML solution. And pretty quick, I think

that they got into the deep learning era. It was, I think, in 2014 or something

So it was pretty early, not from research point of view, but from industry point of view

And it started to get into this field and understand it well and better

I have to say that back then we actually implemented the algorithms for the final solution in the real time

and all of that. So it also provide me software skills and I think it is really important

Going forward, I started to get more and more experience and became a technical leader

Specifically, I got into the field of, there was many cameras around the vehicle

So how to fuse it together, to all the fields of sensor fusing, et cetera

it was really interesting in my eyes. So I started to lead this field and at some point I became a team leader back then

And it was a small team that actually started from one person into two and we were three

people at the end of the day. At that point, a few years ago, I moved to Microsoft

started to develop a new platform that related also to the autonomous vehicles world and it

has like two parts the engineering part and the machine learning part and over then I led the

machine learning block or part you name it and I got more experience not only in machine learning

and algorithms, but also how to get it, how to talk with the product managers, how to

talk with customers, how to build a very big idea, you know, going forward

I also did some acoustic related algorithms and models in Microsoft as well after that

And today as I mentioned at the beginning I am in Surface in the position of data science manager Amazing What a journey Yeah

Really interesting. And it kind of illustrates that nobody on the show

studied like the AI thing are doing actually what they've studied for now

Especially in the field of AI. It's sometimes related, but it always changes along the way

very interesting. But by the way, I see it's really related because it always was algorithms

And it doesn't matter if it's in the biomedical field or in, you know

surface today, so computers or Azure for cloud solutions. It doesn't matter

But the main thing, I mean, for me, it's the algorithms and the data

Perfect. Well said. So it seems like your job is all nice and fun, but I really want to know what is super

annoying about your role in AI. Yeah, so there are some things that are annoying

I would call it challenging. Challenging, that's a good word. I think that the most challenging and sometimes annoying part is actually not the job itself

I mean the AR and algorithms part, but it's more the interface with people that know, not

necessarily familiar with AI. And I'm not talking about how it works under the hood, that's fine

But people think that AI or algorithms can use as an API

This is a very, you know, software oriented point of view. And we got really to a really good place today with algorithms and especially with deep learning

but it's still not, you know, general AI. You cannot take like a model, use it out of the box and just use it whenever you want

I mean, there are publicly known models that people can download from the internet

And this is a great starting point. But if you want to do a real product that work well with high accuracy, algorithmic quality

So you need to use a specific to develop actually the algorithm for a specific domain with a

specific sensor. And as more it will be more specific and related to the target product, it will be better

And sometimes this misconception of using it as API, as I said, can be really hard and

difficult to explain to the other part. Yeah, that's a good point because you work very much in that bespoke machine learning

part. We talk a lot of different people, they work in all sorts of different things, they're

working on all sorts of different problems. Some problems are higher level problems, and

as you said, some problems are very, very bespoke. So it's interesting to hear that perspective

so thank you for sharing. Already, we only have around 10 minutes left. So what Ben

We've spoke about this in our prep sessions. What's coming up next is something called our quick fire round

So in our quick fire questions round, we have six questions. Myself and Hank will alternate asking you these questions

And one thing that we ask is we'll keep the speed up

So we ask if you could answer with either a word, a phrase, or no more than a sentence if you can

And then we can dig into any of your interesting answers as we see

Does that sound okay, Ben? Yeah, I'm starting to count. I know, yeah, I was gonna say, it's all good

Me and Henkel control how many questions you just think about the answers

So let's get going with that. So the first one, Ben, is very, very simple

is more about your history. So what was your first computer? Well, it was actually 2286

It was in, I think, in 1992. Nice, very, very cool. So what program language was used

in the last project you worked on? Python. It's always Python. I'm surprised

It was easy. What is the most useful thing you've learned in AI? What

I couldn't hear you. What is the most? What is the most useful thing

So is it a tool? Is it an approach that you've learned in AI

You always need to look at the data and understand the data well

Very good advice. What is your favorite event on the AI calendar

This is a difficult question because actually I need to get to more events

The last event that I went actually face to face was two weeks ago

It's called IMVC, Israeli Machine Vision Conference. It was really nice after two years of COVID

Can I imagine? Yeah, of course. Now, did they do, was there hybrid stuff

that something people can catch online fully fully in person yeah that's cool

that's cool it's a thank you for sharing and what area is on your list to skill

upon next in AI what's like a new area of AI you're what you want to learn I

I think I very like the area of time series and I developed in this area, but more with

the classical approaches and not with deep learning approaches. And I really want to get into it and to know it better

That's a good one. That's an interesting one. Cool. Final question. You touched upon it already a little bit in your introduction, but now we want to hear

some details because we have some time to get some more details. So you're allowed to

answer like a little bit more than one sentence on this one. So what was the first

thing, project or skill you built in AI? I think I want to go for project

What was the first thing you built? Yeah. What was the first thing you actually made

But AI, not algorithms in general. Because it Yeah, this is difficult

I mean, I can answer both, but the first major project was actually, I mean, it was with a

bit of machine learning and mainly algorithms. Okay do that one Yeah tell us about that one Yeah So actually it was at the end of the bachelor degree my final project

It was related to a sonar actually. And we tried to detect a human in a room environment using sonar

interesting point about it that I was a part of a lab that it was actually a bat lab. They actually

had the bats on it and they were inspired of their sonar capabilities. But we did it with a

real microphone and a speaker. It was a great project because I actually did it from A to Z

starting from the hardware, try to get everything to work together, to define and understand

how it works, to write both to implement the algorithm and to think about it and to define

like the mathematical approach that we are going to use. And it was also with machine learning

because we all we detect and classified uh the different you know activities of the person

uh so i also use machine learning it was classical machine learning and if i'm looking at it today it

was so you know not a lot of data i mean like 20 people or something like that but you know

So it was research, but it was a really good project and we got actually a paper on it

on IEEE, which was really nice and we had the patent that we applied

So it was also a successful project and I learned about a lot in it

That's amazing. Goodness me, academia and education and research are very, very good for that

end to end you truly understand the full problem that you're trying to solve. I think when you said

that I was like gosh I think the last project I've worked on to where I had control of everything that

happened was probably my research project so yes it's a really good perspective yeah thank you for

sharing that one. Yeah yeah a very large project was in the thesis my thesis work so it was much

much larger it was actually in the area of speech but it was you know in during my way not at the

beginning of it. Yeah so then was there multiple different people working on you with on that same

project with the speech stuff? Yeah it was a different person but it was you know much more

research it took much you know lots of time and we have you know really a massive mathematical

formulations and development behind the scenes incredible goodness me audio speech sonar

vision you've really covered so many different areas and then we wish we could spend more than

30 minutes with you. For those who can, do join us in the cafe afterwards. Continue to ask Ben

about his amazing career, about the amazing things that he's worked on. Or if you have any questions

about how he learns, how he got into it, it's a great place to be. So if you go to aka.ms

slash a bit of AI dash cafe, you can chat to us a little bit more. But Ben, with that, we will have

say goodbye to you so thank you so much for joining us we really appreciate you sharing

your perspective your journey and what you do day to day with us thank you for joining us

thank you Amy thank you Hank it was great to meet you thank you no problem all right we'll catch you

later. Very, very quickly. Hank, a couple of announcements we have on our list today

We had if you missed Microsoft Ignite at the start of November, you can catch all of the AI

sessions at akms.ignite-ai-sessions. There is all sorts in there. Store it in what

what they call your back. And you can, it's not a physical one

don't worry, you don't need to wear it. You can store your sessions in there

and you can check them out as and when it suits your agenda

Hank, do you wanna tell us a little bit about the AI show this week

Yes, so I just read the announcement. It's going to be on Friday, so tomorrow

And it is all about managed endpoints in Azure Machine Learning. And they're going to cover and talk in depth about that feature

So that feature enables you to like kind of quickly make a deployment of your model that is wrapped into a container

But these managed endpoints for you can automatically scale and you can deploy multiple, you can have multiple deployments in one endpoint

So, and it can route traffic to those different models running in those individually scaled containers

So, if you have a new model, you can say like, okay, route like 10% to this new model

And then you can see the results and performance of those models, all in application insights

which is like super cool. And then you can just deploy that with like one CLI line

So I'm curious. It's really... Mm-hmm. Sorry, again. Yeah, so I'm curious

Oh, sorry. So I'm really looking forward to that episode of that is one thing I do most of the time

like getting models in production. So this is a great tool. I can tell you're very excited about this episode

From my point of view, that just seems like the area that really scares me the most about

deployment and containers and compute and load balancing and all of that kind of thing. So

A-B testing dealt with for me. Sounds like it's exactly up my street. So, thanks, Hank. So, yeah

if you go to aka.ms slash A-I show, all one word, you will be able to check out previous episodes

as well as find the episode coming up. But that's all we've got time for today. Thank you so much

for joining us. We hope you've enjoyed the show. Catch all episodes of season two as well as

season one which felt like so long ago now at a bit of AI dot show come and join us in the a bit

of AI cafe straight after this at aka.ms slash a bit of AI dash cafe or if you're on our website

it's at the top of the screen and with that thank you so much for watching as always this has been

the a bit of AI show with Henk and Amy we'll see you soon bye