📷 University of Toronto
Journalist: Jeryn Anthonypillai
Jeryn: Hello, and welcome to SciSection. My name is Jeryn Anthonypillai and I'm a journalist for SciSection radio show broadcast on the CFMU 93.3 FM radio station. We are here today with Dr. Andreas Moshovos, a professor for electrical and computer engineering at the University of Toronto. Thanks so much for joining us today.
Dr. Andreas Moshovos: My pleasure. Thanks for inviting me.
Jeryn: So to begin, would you like to give a quick introduction of yourself, like an overview of your academic background and maybe your interests?
Dr. Andreas Moshovos: Okay. Yeah, let's start with the academic background. So I studied in Greece. I did my undergraduate studies there and my masters in computer science, in around 92. I moved to the USA at NYU, where I started studying for my Ph.D. there. I stayed there for one year, and then I moved to the University of Wisconsin, Madison, where I finished my Ph.D. at the end of 98. And all of these have been in computer science. I started teaching at Northwestern University in Evanston Illinois in the US, but whatever reason I want a more European environment. So we decided to give Canada a try. So I've been lucky enough to be offered a position here at the University of Toronto. It was a great opportunity. So I jumped on that. So since 2000, I've been here at the University of Toronto, it's in computer design at the electrical and computer engineering department.
Jeryn: That's great. And currently, you're serving as a director for the national science and engineering research council and computing hardware for emerging intelligence, sensing applications. Do you want to talk about this council's goal and what you guys do?
Dr. Andreas Moshovos: Yeah, so the type of work that I've been working on is how can we design machines that can execute applications better? And what I mean by that, it might be a machine that gives you more capabilities, you know, think about the graphics that a machine could have back in the early nineties, where you could barely see pixels moving. Let's say very simple things to compare to the graphics you see today. So my career has been about the hardware mechanisms that go into enable all the very creative people that write software, which is the most important thing to be able to deliver more through the software by providing more capable sense. In the beginning, this was about performance only as a desktop for machines, lately in the past two decades, been performance per watt, which allows you to take this computing power and move it into edge devices. Things that you can put on your hand or not in order to optimize them as seen and to give, provide more capability. It has always been the case that if you target a specific application or a specific class of application, you can probably build a machine that's extremely, extremely powerful. That's why, for example, your systems today, you have CPU's, which are the central processing unit that are very good at doing general-purpose computing like Microsoft Word, or do your browser. But then if you go into graphics, you have these super-specialized machines that are for graphics processing the GPU, right? And pretty much every machine today has both. Machine learning has emerged as one application category that's basically has taken everything by storm. Whereas in the past you would have to program into a machine, all possible scenarios that ever occurred so that you can make it do something you want with machine learning. You have at least the impression that it actually learns, right. It can learn how to do that without actually you program every single scenario. So I was lucky enough to be at the University of Toronto and see a lot of innovations happening in my city. Learning early on. It seemed to me that this was a domain for which we can actually try to build very, very specialized machines that will allow you to take these applications and use them anywhere and for anything.
Jeryn: Yeah, awesome. And also your current research at the moment is on highly specialized competing engines for deep learning. I think you kind of slightly touched upon that, but would you like to go more in depth on this research that you're working on?
Dr. Andreas Moshovos: I should have said about them, NSERC COHESA instead I talk about only what I do and what we do. So the NSERC COHESA . Let me go back to that before. It's actually a consortium of several researchers, 24 at the moment, and several communities that do have a base in Canada. And the whole goal is to design systems that allow you to build ever more clever and ever more powerful machine learning applications, most of the researchers in NSERC COHESA are people very similar to my expertise, right. They have very similar expertise, they design hardware, but there are also several people in NSERC COHESA do machine learning per say. And then there are several other people that actually do systems. Systems from the perspective of the system software, as in Linux, Windows and everything that goes in between that allows programs to run that. You asked me something else that I haven't said, can you remind me? Sorry, I got lost now.
Jeryn: Yeah, no problem. So I was asking you about your current research at the moment, which was on highly specialized competing engines for deep learning. So would you like to go more in-depth on this research that you're working on?
Dr. Andreas Moshovos: Yes, so this is about building machines. When they're called to run machine learning applications and in particular deep learning networks, they allow you to run even more powerful applications. Whereas before to run a machine network, you probably admitted their desktop CPU with a lot of GPU. So you can get this thing to work, even work today. You can have even a cell phone run a lot of this stuff. So our goal is to try to enable even more powerful applications of that. So the way we do that is by designing machines that are specialized for this kind of application, they are not very good for doing graphics. They're actually horrible for that. They're not good for running Microsoft Word or running a browser, but are extremely, extremely good at running these particular applications. And the reason we can do that is because if they're very low level at the hardware level, these applications boil down to gazillions of multiplications and additions. So we build machines that can do this extremely, extremely efficiently.
Jeryn: And what do you think was the most fascinating thing you have researched throughout your years? Cause I know there's a lot of new and exciting things that there is to research in this field. So what do you think in your opinion will be the most fascinating?
Dr. Andreas Moshovos: So this is a tough question to answer really, right? Because I was blessed enough to like computers as a kid and through a series of events. Let's say, I mean, some of them are pure luck, right? Events. I got to work on what I like a lot. So from a purely selfish perspective, I am blessed enough to be working on something that I really like and pretty much everything that I've done so far with and the people have collaborated with it. People have collaborated with have been very, very fulfilling for me. This is from my own perspective, right? In terms now of how these are exciting for the rest of the world, well you have to go, unfortunately, what you do, design a car where there's a lot of levels in the direction that you have to follow through to say that something adding should these things end up being products and an actual message that you use, and it takes a while for this to happen. You're going to be able to see applications coming very new to you next to your hand, and a lot of applications that today are even possible now are going to become possible. The most fascinating thing if we want to talk about fascinating, if you can put that term or the work we do is that the machines that we built, actually look at what you want to do and they try to do better than you say. If you are a programmer, you tell me you want to do a thousand multiplications. Well, my machine is good to maybe 500 and still give you the same result. That's the most fascinating thing. So there are machines that actually look through instructions. You give them and only do as much as necessary to give you exactly the same result without you trying.
Jeryn: And I think a follow-up question to that would be, what do you think most people should learn about your research?
Dr. Andreas Moshovos: In terms of the daily life of a single person? How this work has helped them, right. Is that what you're asking?
Dr. Andreas Moshovos: So this is again another difficult question to ask boils down to the following. The fact that you have a computer in front of you, or the fact that you have the car that can probably see the street and you know, the road and drive itself to some degree or the fact that you have a plane they can actually navigate itself and take you to a destination and take actions when there is danger, computers are everywhere. So even though I do not write applications, the reality is that pretty much everything in science, everyday life is being affected by computers, mostly in positive ways. Yes, there are negative effects. So the kind of work that we do in NSERC COHESA, my group is the kind of work that allows these computers to become ever more powerful and enable all of these conveniences or new applications. I mean, drug discovery in many degrees today, it's done first in computers, you know, designing airplanes happen first in computers. So there is a level of indirect. So it is not direct.
Jeryn: Yeah, it's definitely everywhere and in our daily lives. And for our final question, you have won multiple awards for your work so what do you think was the root of your success?
Dr. Andreas Moshovos: Oh okay. First of all, you're very kind, it's very nice of you to talk about success. That's a question that everybody has to look at in the mirror and decide whether they really think they're successful or not. I was blessed as a kid. A lot of stuff made sense to me very easily. So I think that was a matter of luck. Right? When I went to school, things became easy for me. So there was a feedback loop that allows me to work hard and get rewarded for that. I do recognize that this may not be the same for everybody. I was just blessed that way. And there was a lot of hard work that went into where I am today. I don't believe that what I am is so important. Anyhow, I was also very lucky to work with very, very talented people as a student, as a professor now at the University of Toronto, there's a lot of great talent in Canada. And I think we're lucky to be professors in Canada because there are great people we can work with. And finally, there is a lot of luck too, right? There's I can think of several events in my life where it could have gone either way. And I wouldn't say that I had much control over them, right. So let's not forget about that. So there is a lot of hard work and it usually works, but it's not the only thing, working with the right people and being there at the right time. So this is a bit you know, I would like to think that I do have a lot of control in my life, but you only have only so much.
Jeryn: Mhm, and so that brings us to the end of the interview. Thank you again for joining me today and for our listeners, make sure to check out our podcasts available on global platforms.