Blog: Interview with AI Researchers: EP-1 Sai Krishna from MILA, Montreal
This is the first part of the several interviews to follow in covering different aspects of Artificial Intelligence research and development, the shifts that the field is facing across different domains, and we will be covering interviews with the key players across all these research topics.
In this interview, we have with us Mr. Sai Krishna Gottipati who is currently pursuing his Master of Science in Machine Learning at Montreal Institute of Learning Algorithms (MILA) headed by one of the founding fathers of Deep Learning — Yoshua Bengio. He is currently conducting research, advised by Dr. Liam Paul, who heads the Robotics research at MILA.
Thank you for agreeing to do this interview Sai.
Could you give us a brief about your background and what got you interested in AI? Tell us a bit about the different research projects which shaped your admit at MILA, from your undergrad.
I pursued my Bachelor’s studies at IIIT Hyderabad. I joined as honours student in robotics research centre with Dr. Madhav Krishna at the beginning of my 3rd year. I actually started working with him in my 2nd year, but got officially admitted in 3rd year. I started working on various computer vision based problems for our autonomous car. For some approaches, traditional vision/image processing based approaches worked really well, for example: lane detection, but for tasks like traffic signal detection and recognition, they didn’t work well. This clearly pointed out what’s wrong with such approaches: they couldn’t generalize well to different scenarios.
This drove me to explore machine learning based methods and I taught myself via online resources. for example, neuralnetworksanddeeplearning.com helped me to get started. I played around with various network architectures, other non-NN based methods like SVM etc.
I also got opportunity to work on a research paper: https://arxiv.org/abs/1609.09468 where I worked predominantly on key point localization using a deep neural network. I also very briefly interned at Speech Lab with Prof. SuryaKanth and his Ph.D student, which helped me check my basics and taught me various hacks for faster and better training. I also tried to keep myself updated on the latest literature in computer vision and deep learning
2. How did you get interested in MILA? How did you prepare yourself to secure an admit at MILA?
I think one of my friend asked me to checkout Mila. I then came to know about Yoshua and his research and decided that this is where I should spend my next many years. I didn’t specifically prepare anything for Mila. just applied for it along with other grad schools.
5. What all factors do you think helped you get admit at MILA? Can you give us a brief about the interview process at MILA before you’re given an admit?
Prof. Liam Paull and I had lot of research interests in common. I think that helped my admit. And of course, good CGPA, a publication, good recommendations also must have played a role. He asked questions about my projects and what projects I wanted to pursue in future.
6. What are your active research interests now? Can you give us a brief about your work at MILA?
I recently submitted a paper to RAL + IROS on deep active localization: https://arxiv.org/abs/1903.01669 My research interests in general are reinforcement learning, generative models, cognitive AI, biologically plausible learning algorithms etc.
7. How does it feel like, working with Dr. Bengio? What is the research culture like, at MILA? (Collaboration, kind of research, how the people are, how open the research teams are, etc.)
I never had one-one meeting with Prof. Yoshua (Liam supervises me). He always has a lot of interesting questions and novel ideas which he frequently shares during multiple reading groups every week. Most professors attend these reading groups and share their ideas. This leads to collaboration between students of different professors. Scientists from various AI companies also frequently visit Mila and collaborate with students on research projects. If any student has any idea, he shares it on our slack and gets immediate feedback. Most of our queries (for example, an issue with pytorch, relevant literature for my current project, etc.. ) are discussed on slack and are quickly resolved. I think all other academic labs should adapt this culture of openly sharing ideas and collaborating.
8. What do you think is the single most important topic in the field of AI right now, that if advanced could set the precedent for decades of research to come?
Reproducing the working of the brain. Prof. Blake Richards gave a nice talk on current challenges and advancements on this topic at Summer school last year in Toronto. I recommend everyone to watch it. Many top researchers like Yoshua, Hinton, and a few companies are actively working on it for many years.
9. Research, especially in AI — is now almost exclusive to industry and academia. With most top line researchers being poached by industry from academia, where do you see the future of AI research, especially from the POV of mentorship?
I am not sure — most of them still maintain their affiliation with University, so I think its okay. collaboration between academia and industry is necessary for a proper collaboration between talent and resources.
10. Research as a field is more exclusive currently, especially in the field of AI. We need as many people as we can in the world of research, actively advancing the different fields, including AI. This exclusivity to academia and big industry giants — how do you think this can be overcome and someone interested in research get into the world of research, especially AI?
I don’t think it’s exclusive. all the publications are available for free for anyone to read, most of the codes are open sourced.
11. What all skills do you think are important for a researcher in general? How has your experience at MILA shaped you in terms of those skills?
Intuition, math and coding. In my last 20 months at Mila, I was never really under any pressure to deliver something before a due date. I should thank Liam and in general the culture at MILA for allowing me to do whatever I want. This is how I got interested in various topics like learning in the brain, causality, cognitive AI etc., even though my primary focus and thesis are supposed to be on robotics.
12. What are your future plans? Are you going to pursue PhD? If yes, what do you plan on doing after your PhD?
I accepted a position as a machine learning researcher in a stealth mode startup in Montreal, associated with MILA. I might return to PhD at MILA after few years.
13. Which company other than the top ones (Google, Facebook, Apple, Amazon, etc) do you think is doing the most interesting work in AI right now?
There are few startups which work on fundamental machine learning and learning in the brain. They don’t have a revenue model, but it’s interesting to see if they achieve any break through.
14. What is the funding package provided to MS and PhD students at MILA? What all does it cover?
MS students are given a stipend of $22K per year. Ph.D students are paid $25K/year. sometimes they pay the entire tuition. or, you would have a nominal fees like $2K per trimester if you are taking courses. Sometimes, they fund your expenses to conferences or summer schools.
15. How is life at Montreal? What’s the city like? What are the monthly expenses like, at Montreal?
It’s a great city. It’s not noisy like most US cities. The public transport is pretty good. It has lot of beautiful public parks and lakes. It’s the best. rent is about 600$, other bills (transport, mobile, laundry, internet etc.. ) will be about 200$, food will be around $400-$1000 depending on whether you cook or eat outside.
16. As part of MS in ML by research, how is the course structured and when do you start your research work usually?
You just need to take a total of 4 courses (over 2 trimesters). The course work isn’t heavy, so you get to read lot of research papers during your first 2 trimesters also. That’s pretty much the beginning of proper research.
17. What are the opportunities for someone who decides to finish MS (research based) at MILA and decides to go for a job?
Many startups (in both Canada and US) will be interested to hire you.
MILA, Montreal Institute for Learning Algorithms is the world’s leading Deep Learning and AI research lab located in the heart of Canada — Quebec, Montreal. Dr. Yoshua Bengio along with his cohorts of professors and students from UdeMontreal, McGill University, Poly, HEC, and UBC, has created this lab to focus on bringing the next generation of Artificial Intelligence into life. Professors as well as students from MILA regularly work with companies such as Microsoft Research, Facebook AI Research, Google Brain, Deepmind, etc., and startups such as Element AI (becoming a unicorn, co-founded by Dr. Bengio himself) in surpassing the abilities of AI that are known to the world day after day, setting brand new benchmarks for what’s cool and what’s meaningful.
You can find more about MILA here.