The Artificial Intelligence (AI) Residency Program is a one-year research training position designed to provide hands-on experience with AI research at Facebook. The program pairs AI residents with Facebook researchers and engineers to collaborate on research problems of mutual interest. Together, the team devises deep learning techniques to apply to exploratory areas and big industry challenges. Participants are also encouraged to collaborate with others across Facebook AI.
“This unique program is geared toward anyone who has an interest in working in AI and getting more practical experience in a world-class lab,” says Tony Nelli, Program Manager for AI Residency. “You don’t need a degree in machine learning, or even in computer science. We’re interested in applicants from a wide range of backgrounds, and we hope to make careers in AI more accessible through programs like the residency.”
To learn more about what it’s like to be an AI resident at Facebook, we sat down with current AI residents Eric Mintun and Diana González. Mintun studied string theory and quantum gravity during his PhD at the University of California, Santa Barbara, and then spent three years as a postdoc at the University of British Columbia (UBC) in Vancouver studying similar subjects. González obtained her master’s degree in science and computer engineering from the Institute for Research in Applied Mathematics and Systems at the National Autonomous University of Mexico (UNAM).
In this Q&A, Mintun and González tell us how they learned about the program, how they found themselves at Facebook, what they’re currently working on, and where they see themselves after their residency. They also provide a few words of encouragement for those considering applying.
Q: Can you describe to us the journey that brought you to Facebook? How did you hear about the program?
Eric Mintun: I had been fascinated by the AI boom since the end of my PhD, but my research had been very analytical and not computational, so there didn’t originally seem to be much of a way I could contribute to the field. The postdoc above me at UBC went to the Google AI residency the year after I arrived. I joined him and several other physicists who were interested in AI in a journal club where we discussed theoretical results in AI. This group included Sho Yaida and Dan Roberts, current and former research scientists at Facebook who had encouraged me to apply to the program.
In addition to this journal club, I spent around a year studying basics and working on a side research project with a professor in the computer science department at UBC to prepare for the residency.
Diana González: I was studying for my bachelor’s when I first learned about AI and quickly fell in love with it, so I decided to do a master’s where I had the opportunity to study different branches of AI such as vision, creativity, data mining, etc. For my master’s thesis, I worked on a task called visual storytelling, where both computer vision and natural language are needed. Since then, I’ve become really interested in teaching machines to understand how we communicate.
My thesis supervisor encouraged me to apply to the program, and even though I was afraid of being rejected, I love research so much that I decided to give it a try. I’m very thankful that my supervisor encouraged me and gave me the confidence to apply.
Q: What drew you to Facebook’s AI Residency program?
EM: There is a large group of people here working on a very broad set of topics. While my background is in theory, I wanted exposure to the more practical and application-driven approaches to problems. Second, Facebook has some of the best researchers in the world, and it’s great to have the opportunity to learn from them.
DG: I knew the program would help me improve my research skills by being part of large projects that provide solutions to real-world problems. Also, getting the chance to deepen my knowledge and at the same time be part of something that can impact a lot of people in their daily lives is an amazing opportunity. Facebook is the perfect place to do this because you can apply solutions to your problems immediately. For me, I’m all about the immediate impact. In academia, it’s a lot slower.
Q: How would you describe your experience as an AI resident so far? Are there any moments that stand out?
EM: The residency has been great so far. The cohort of residents is pretty tight-knit and a fun group of people to talk to about research or anything else. The vision team that I’m on is supportive of each other and working on all sorts of cool things. My mentor Saining Xie is extremely responsive, excited about what we’re working on, and fun to collaborate with. One thing that has stood out is that we’ve had the opportunity to sit down with a few authors of seminal papers who work here and learn what they were thinking that led to their results. Also, we’ve had a few opportunities to talk with the senior leadership about the company’s vision for AI and how the residency fits into that.
DG: My experience so far has been really exciting and nourishing. The reading meetings we have are one of the best parts of the residency. At these meetings, a researcher explains three to four of their papers to us. It’s so cool because they’re talking about papers that they worked on themselves, and they can give you the background knowledge and insights. Because the residency just recently started, I haven’t been to any conferences yet, but I’m really excited about attending those. My favorite part is definitely working alongside the amazing researchers here at Facebook and sharing ideas with them, as well as learning from the other residents.
Q: What are you currently working on at Facebook? What have you been learning?
EM: I am currently working on equivariances and symmetries in neural networks. The mathematical structure of group theory allows people to build neural networks that are provably robust to a certain set of transformations. We hope that by backing off the mathematical rigor but using similar intuition, we can expand these methods to approximate robustness to a much larger class of transforms. As an independent side project, I am looking at analytical solutions to very wide neural networks. I haven’t settled on a concrete project to pursue in this direction yet, though.
DG: For my AI residency, I’ve had the chance to work very closely with incredible researchers like my mentors, Y-Lan Boureau and Emily Dinan; Eric Smith; and other members from the ParlAI group that focus on dialogue systems. My main research focus is the analysis of people’s preference on the talking style of their conversation partners. To get there, I’m also working on finding more successful ways to do style transfer on text, which would have good applications in dialogue systems (emojis/without emojis, formality and politeness, etc.).
Q: For someone who might be discouraged about applying to this program or on the fence about applying, what would you say to them? Any advice/words of encouragement?
EM: Before applying and coming here, I had zero formal training in AI or even in anything computational. The other residents come from a pretty diverse set of backgrounds, including AI research, pure software engineering, industry work in finance, research in neuroscience and physics, and more. If you are interested in AI research, it’s worth applying regardless of your background.
DG: Do it! There’s so much to gain and so little to lose if you apply. It may feel scary, I know, but it’s so good to challenge yourself to go for a little bit more, and it’s very likely that something amazing will come out of it!
Q: What’s the plan after your residency? Where do you see yourself in the near future?
EM: I’d enjoy staying here in FAIR’s two-year postdoc position, if possible. Also, during my time here, I’m hoping to get a better sense of how the engineering side of AI operates, and I might be interested in pursuing that.
DG: I would love to continue my journey in research and pursue a PhD where I can continue working in machine learning and natural language. I want to keep learning and contribute to the research community.
Q: If you could send a message to your past self from five years ago, what would you say?
EM: I’d tell myself to start getting involved with all of this earlier, and to actively seek out like-minded people in my field. It’s half a matter of learning and half a matter of finding someone who could steer you in the right direction. There are a lot of resources online these days, including standard ones like Coursera.
Several other theoretical physicists were already then becoming interested in the field and would have been great to talk to, but I wasn’t aware they existed until I was in the same lab as one. So I would tell myself to network ahead of time. There are probably more people in your field who are interested in AI than you might realize.
DG: Keep working on the things you feel passion for and, above all, don’t quit, because greater things will come. And also, whenever you can, share your knowledge and experience with others.
Applications for the 2020–2021 AI Residency program will close on January 31, 2020. For more information and to apply, visit our program page.