July 16, 2018
Interview by Anjali Bhardwaj
Deeplite created an AI-Driven Optimizer to make deep neural networks (DNNs) faster, smaller, and energy-efficient from cloud to edge computing. In April 2018, the team got incubated by TandemLaunch; now a few months later, the team has won the C2 Montreal 2018 Emerging Entrepreneurs Contest amongst other awards. We chatted with Davis and Ehsan, Co-Founders of Deeplite who met and assembled their startup through our Entrepreneur in Residence Program.
Can you tell me a little bit about yourselves?
My name is Ehsan Saboori. I did my Bachelor’s in Computer Software from Ferdowsi University of Mashhad, and my Master’s in Information Technology from K. N. Toosi University of Technology both from my hometown in Iran. In 2011, I came to Montreal to do my PhD in Embedded Systems at Concordia University. During this PhD, I did two internships, one at SAP in Montreal and the other from Microsoft in Vancouver. After I graduated, I began working at Morgan Stanley for about a year and a half. I then left Morgan Stanley to work at TandemLaunch.
My name is Davis Sawyer. I was born in Canada but I grew up in Houston, Texas. I studied in Boston at Northeastern University and afterwards, I started working in oil and gas for a while. I found that exciting because of the exposure to the power and data industry. A lot of people say data is the new oil and I have to agree because my first job was to make a database for an oil company.
I later moved on to biotechnology and worked in a company where they manufacture a therapy for Crohn’s disease. It was very purposeful work, but I always wanted to go back to Canada to innovate back at home. I knew that Montreal in the last couple of years had really taken off in AI and is world-renowned in that way. Last year I heard about TandemLaunch and I was finishing up my contract in Cambridge. I had met a couple of the founders and a couple of companies like SPORTLOGiQ, and I was hooked. I liked their process and I liked the team, so I joined last year in September and started Deeplite.
So what is Deeplite?
Deeplite started on the idea that right now most AI starts on the cloud, in large data centers. That is fine for some applications such as Google Photos but right now the best way you can help people is actually on edge devices. So how can we actually use AI on these low powered systems like smartphones, smart cameras and vehicles? Before Deeplite there was not really a good way that you can use deep-learning and AI on these devices. And so Ehsan had actually thought about this, maybe last year when I first met him when he pitched at TandemLaunch. I thought that it was the coolest thing I’ve heard and so we teamed up around this idea of how we can make deep learning applicable for these other devices.
Deeplite is an optimization software that helps you use real-time AI applications such as computer vision. If you wanted to say, look at your crops in a field and you don’t have internet connectivity to see if they are healthy or not, well now you can use AI in that device. For us, it was about making this technology affordable and accessible but also scalable because right now there is a huge demand for this and so how do we do this without draining our resources in terms of energy and also keeping a levelled playing field in a very kind of democratic use of AI. So that is the company’s mission I would say.
In terms of the technology side, if I wanted to put it in a nutshell, we are optimizing deep neural networks to make them applicable for low end and tiny devices. To achieve this, we hire different expertise from different fields to work on that. Our main goal is to make this process automatic, without any human interaction.
Right now, if a company wanted to do the optimization work you would have to hire an expert who would just be exorbitant for the price tag for a lot of companies. A lot of companies want to use AI as a tool to start their businesses, but they don’t have the cost structure or the internal expertise to do so. That is why Ehsan mentioned that it’s to automate the process of making AI on your smartphone or on your vehicle so you don’t need some high-end or expensive expertise. That is a part of the affordability side. Also as Ehsan mentioned, Deeplite bridges the gap between programming languages and processors. Because of that you need a pretty interdisciplinary team and so some of our team is on the hardware side and they know all the hardware and some of our team are more programmers so software side, and it is cool because you see this interaction between different backgrounds.
I think you really need to be happy with your work and you really need to enjoy what you are doing. The odds are already against you, so why not enjoy it.
What was the biggest challenge you went through building Deeplite?
In the TandemLaunch model, we have to prove that the technology and the market are feasible, and not only that but you need a prototype. But really one of the most challenging things for us was to prove that the technology works.
Yeah, I think for a deep technology it is really a chicken and egg problem because you want to de-risk and say that it is feasible but it also takes lots of resources to make that happen. So, you want it as soon as possible but also in a sustainable manner. One of the things on the industry side on the market validation is that AI handles workloads that would, for example, detect a pedestrian and so, that is a huge factor in terms of robustness. It is not just a matter of the AI only detecting that there is a pedestrian, it needs to be real time, it needs to be happening as soon as possible and so for us to get it to the market and to help the market we have to be doing these mission-critical things and so it’s a big hurdle for us as a startup. But it’s something that we are definitely tracking down.
How did you originally come up with the technology?
In terms of the optimization technique I was looking at different universities and connected with different professors but one of our main focuses was at Brown University because they were working with different aspects of this optimization which really could feed into different parts of our framework, and I contacted Sherief. The first time I talked to him I told myself, “Yes, this is the guy I have to work with.” And we talked to him and began developing the technologies and the prototype.
I remember once I was talking to Ehsan and I asked him, “So, what have you done so far?” And he had a big stack of papers in front of him and he said, “This is what I read.” And I tried to catch up on my reading so we can be on the same page, literally. He did a really big literature review and we felt like Brown University had the right person and the right technology. They had the most promising one.
Now when I look at stuff in the world, I see that it is not a miracle. Someone worked their ass off to make this happen so now I don’t take things as much for granted anymore.
Going back to the biggest challenge you guys had to go through building Deeplite, what did you learn going through that?
Don’t give up! Because my creation phase took nine months instead of six months, it was a longer creation phase than normal. For me, when I started working on this project and had gotten the idea from the university we needed to make sure that the technology is working and the market is great. There will always be an obstacle that you will need to push through to move forward, I think really the most important thing I learned was to not give up and keep pushing to get it. It was possible for me to give up after six months and say that it was not working. The last three months was the toughest part for me, to push through and to make it happen.
I remember one day I saw Ehsan at his desk and he was looking at his old mug from Morgan Stanley and, I was thinking, What the heck is he doing? And I asked him, “What’s up?” and he said, “You know this mug reminds me of how much I didn’t like it there, so this has to work.”
Did your interest in tech start from a young age?
On my side yes, I have always enjoyed programming. I got my first computer in 1998. My first program was to just print out the numbers from 0 to 100 in different colours in Q-Basic. I was in 6th grade. So yes, I was always like this.
I definitely took technology for granted a lot in my life. It was in the last job that I used technology for statistics and probability. We were working on predicting the safety of a drug and I tried using a very basic implementation of a neural network and based on our conventional accuracy and the new accuracy. That was a kind of moment for me realizing that there is a lot you can do with technology. Especially working with Ehsan, the appreciation I have for how things work and how people get stuff done, it is mind-blowing. Now when I look at stuff in the world, I see that it is not a miracle. Someone worked their ass off to make this happen so now I don’t take things as much for granted anymore.
I remember one time I watched this one-year-old and they had this print book. They were trying to swipe the pages. It’s just a sign of how quickly humans adapt to new things.
What quality do you think is necessary to become an entrepreneur?
One thing you realize when working in the industry side and even the technology side is that you can literally learn from everybody. Especially in entrepreneurship, you don’t know all the answers. And the more people you talk to the more advice they share and more often than not they know more about something than you do. And when you have a place like TandemLaunch you have so many people with so many different backgrounds altogether and you can really accelerate the way you learn. For me, it was realizing that if you just pick around a little bit you can learn so much from everyone else. And the second thing is, we call it an “Irrational Exuberance” but just believing in something even if all the facts and all the current information doesn’t point at something. You have to make a bet in something, in what you believe in, what you think you can do because the market will tell you different things, the investors will tell you all sorts of things, even you will tell yourself a bunch of different things but you have to have that kind of conviction for sure. And I have seen it from the successful founders here and it has always inspired me.
I think as Davis mentioned, you have to believe in what you are doing and you have to enjoy what you are doing otherwise you will not be able to tolerate it. There is a risk and there is a pressure of completing things in a certain time period from everybody. I think you really need to be happy with your work and you really need to enjoy what you are doing. The odds are already against you, so why not enjoy it.
What is your commentary on the future of AI?
I think that right now we are only in the beginning and I think in the future we can solve so many problems with AI and it can help humanity with so much. But now there are so many negative ideas around AI, like how AI will stop the human race but I don’t believe in those.
And with every kind of industry, medicine, etc. It jumps and it doesn’t crawl. I think right now as Ehsan mentioned, we are in the beginning phase but more often than not in these kinds of domains you have so many people excited about something, so many people researching something and new papers coming out every day and there is that incremental progress in these industries and then all of a sudden that floodgate opens. It is impossible to predict what it will be but you can always guarantee there will always be that kind of innovation just by the nature of it you have all that randomness, all that passion blended into one thing it makes breakthroughs regardless.
Yeah, it is a revolutionary technology, much like when the computer first came and how it impacted every single person’s daily life. And right now you can’t imagine a world without computers, it’s the same for cars. I think in the future when we have more advanced AI we won’t be able to know what it was like when we didn’t have AI.
I really do think AI will reach a point where it is kind of the expectation. I remember one time I watched this one-year-old and they had this print book. They were trying to swipe the pages. It’s just a sign of how quickly humans adapt to new things. And as Ehsan mentioned, at one point it will definitely be woven into daily life for better or for worse, but it is obviously hard to say. I also think another thing is understanding, I see a lot of companies and governments trying to increase this kind of public awareness, not just in the PR sense but an actual understanding of the technology. With how revolutionary of it also comes a bit of mysticism and when you go to conferences and when you go to presentations, both academic and non-academic, you have that kind of element and it is exciting. I was in Germany two weeks ago and a guy presented on the hyper-growth, the weak AI that is detecting an object and the super AI which is thinking and reasoning, and multitasking, and that line is super non-linear, it is not at all step by step in my opinion. I think understanding the reality of it is huge and something we keep learning every day.
Thank you for your time!