tl;dr: Thank you for your interest in working with me. I receive many emails regarding joining the Adaptive Agents Lab. I wish I could answer all of them in detail, but unfortunately I cannot. Please don’t take it personal! Instead, I present you with a FAQ. Hopefully my explanation here answers most of your questions.
Graduate Students: It is likely (~60-80%) that I won’t recruit any new students for the academic year 2024. If you want to apply anyway, and if you think that there is a very good match between our interests and you have the right background and skills, apply to the Department of Computer Science, University of Toronto and mention my name in your application form and SOP.
Undergraduate Students, Interns, Visitors, Postdocs, etc: I won’t take any student or trainee for the academic year 2023-24.
Graduate Students
Q: Will you recruit any new graduate students for the academic year 2024 (Fall)?
A: There is a good chance (~60-80% – subjective Bayesian probability à la E.T. Jaynes) that I won’t recruit any new graduate students this year. If despite this you still want to try your chances, please mention my name on your application.
Q: Will you become my supervisor or help me in the admission process?
A: Admission is granted by the department, and not by me. I will have a say on whether I am interested in a particular applicant or not, but that is all! I will decide on whether or not I am interested in working with a student only after seeing all the applicants and looking at their files closely. This will happen in late Fall and early Winter (December-February). Unless you are really an exceptional match (you have already published 2-3 very good papers in my research area in top venues such as NeurIPS or ICML), I cannot really provide a more concrete answer before then.
Q: How can I ensure that you look at my application?
A: There is a field in the application where you can specify your interest in working with faculty members. If you write my name, I will look at your application.
Q: Do you mind if I contact you meanwhile?
A: No, but it may not be needed! FYI, approximately half of all offers initiated by me in the past 4 years were to students who hadn’t contacted me beforehand. I cannot definitely claim their emails didn’t have any effect at all (I need more data!), but I believe sending email is not a must for me.
I will go through all applications that mention my name, one by one, no matter whether you email me beforehand or not.
If this isn’t comforting enough for you, and if you think we are really a good match, feel free to send me a note. But please don’t take it personal if I don’t reply back. I probably won’t, and it is not because I enjoy being rude, but because if I reply back all emails, I will lose a lot of time that I could spend on research and working with my students (so, please don’t follow up if I have read your email; I have probably read it). Just make sure to mention my name in your application. That is the best way to guarantee that I look at your application. If you want to contact me regarding admission nonetheless, please write Dawn of Machines in your subject line. This indicates whether you have actually read this page.
Q: How should I know if we are a good match?
A: Generally speaking, I am looking for a student who is an independent thinker, mathematically mature, organized, has ML/RL background, has some research experience, and knows how to write good code. Please take a look at my research and papers to get a better idea of my research style (and don’t look only at the last paper as it may not be the representative of my work).
- Mathematics: My research is theory-inclined, so you need to have a strong background in math and computer science, and enjoy approaching ML problems from the mathematical perspective. You need to feel comfortable working with probability theory, linear algebra, calculus, etc. You also need to have experience proving theoretical results and be mathematically “mature”. Take a look at some of my papers to get a better feeling about what kind of math I am taking about.
- RL and ML background: You need to have some background in ML and RL. For example, you need to have taken some relevant courses or read a good part of a standard textbook.
- Research experience: Research experience is very desirable. Good MS students have some research experience in their undergrad. I expect someone who applies for a PhD program after their MS program (as opposed to direct PhD from their undergrad) to have prior experience in conducting an independent and original research and writing a manuscript in ML/RL-related topic, and ideally some publications in good venues (such as NeurIPS, ICML, etc.)
- Programming: You need to be able to implement your ideas quickly, clearly, and efficiently. If you have an online code repository, make sure to include it in your CV.
The standard for a PhD student is higher than for an MS student. For example, an MS graduate needs to have research experience in areas close to what I do.
Q: What is the probability of my admission?
A: It is a conditional probability after all, and part of the condition is you, and another part is others in the pool. I cannot honestly give you any good estimate. Just know that the CS department at U of T receives a lot of applications, and many of them are strong. Unfortunately our resources are limited, so we can only accept a limited number of students. Personally speaking, I expect tens of students show interest in working with me. This is my privilege, and I acknowledge it. Unfortunately, I can only recruit a very limited number of students per year, perhaps only one or at most two. As a result, the prior probability of being accepted to work with me is low. Note that we have a relatively large ML and AI group here, and you should consider them when you apply. This increases your chance. But overall, the chance of admission is not very high. I understand this is unfortunate and many great students may not get admission at the university they wish. You need to know that the admission process can be quite noisy, so if you don’t get admitted, don’t consider it as a judgment on you or your ability.
Q: What will happen after I apply?
A: Many things happen at the department and university level. I describe my actions in some more detail: After all the applications are in (early December), and the pre-screening phase by the department is finished (middle/late December), I go through all those who have mentioned my name. I will read their applications (look at their CVs, transcripts, reference letters, SOP, etc.), and will give them a rough score. This happens in late December or early January. I will shortlist a handful of them, start the conversation with them through email (early/middle January), possibly have a video chat with handful of them (late January or early February), and perhaps even ask them to give a short 10min talk about their current research (if they have done some previous experience). Based on these, I will express my interest in working with some of them to the department. The admission committee considers my note in making their final decision.
Q: Why don’t you admit more students?
A: Two of my main resources are 1) time and 2) funding. Both of them are unfortunately bounded. Even if I have an unlimited amount of money (which I don’t), I don’t have unlimited time to advise and guide my research group. Having more students means that I have to spend less time with each of them, which reduces their productivity, and is not fair to them. Moreover, it means that I cannot get involved in their research as much as I like. And I like to be involved in their research – that’s the point of being a researcher after all.
Q: I am an undergraduate student. Should I apply to the MS or PhD program?
A: If you are an international student, do not apply to the MS program, but directly apply to the PhD program (why? keep reading!). If you are an undergraduate domestic student, apply to whichever you want. My preference is that a domestic student starts as an MS student, but comes with the intention of working towards a PhD. The reason is that the PhD program allows more time to do good research and get deep into a topic. An MS student and I work for 1.5 year together, and if we are both happy, the student transitions towards PhD. But if you really don’t want to do a PhD, that can be discussed, but please be upfront.
The reason I am suggesting international students to not apply to the MS program is because of how the funding system at the U of T works. There is not much subsidiary for international MS students, so they would be very expensive for a supervisor. Unless an international MS student has an extenal source of funding, some (most?) professors decide not to recruit them, and prefer an international PhD student.
Q: Should I apply for external funding?
A: Yes, if possible. If you are a domestic student, I encourage/expect you to apply for government fundings such as NSERC and OGS. If you are an international student and there are similar scholarships in your country, apply for them. Note that having a funding is not a guarantee for your admission, but it removes some constraints. What is most important is that we can be a good match.
Q: Who else do you recommend if I want to work on AI/ML/RL?
A: There are many excellent researchers around the world working on AI/ML/RL. Let me focus on Canada first. The Machine Learning group at the department of CS has several faculty members, who work on a variety of ML-related topics. There are other PI at DCS working on other areas of AI too. Take a look at researchers affiliated with the Vector Institute, Amii, and Mila (To the first order of approximation, these are AI/ML/DL-focused institutes with close ties to universities. Most of their members are affiliated with universities). There are several other excellent professors at other Canadian universities too, which are not part of these institutes. For the list of RL researchers at the global level (with a focus on North America), take a look at this spreadsheet (maintained by Philip Thomas).
Undergraduate Research and Internships (Graduate or Undergraduates)
[Currently (Summer 2023) I will not recruit any new student or trainee for Fall 2023 or Winter 2024, as I am close to my bandwidth.]
I recruit undergraduate students who are interested in doing research as well as undergraduate or graduate-level interns. You will probably work closely with one of my PhD students.
- University of Toronto: I usually have some projects available, and I often announce them through the department’s Undergraduate Summer Research Program (UGSRP) (early Winter), though I may occasionally have some projects off-cycle too. Please send me your CV, transcripts, and your availability (when you can start, how long you can work, and how many hours per week you can dedicate to the project).
- Outside Canada: I may recruit an intern through the Vector Institute’s Internship Program.