Advice for Prospective Graduate Students

Advice for prospective graduate students

Every noble work is at first impossible. --Thomas Carlyle

What students am I looking for?

  • Independent thinkers. How can you prove that you are an independent thinker? Example: You have published a paper that builds on your work. However, not all papers count. If you did some experiments in a paper and is merely presenting the results, without trying to teach a lesson that generalizes beyond the scope of the experiments, this does not prove that you are an independent thinker. The lesson should be novel. We all know that genetic algorithms, given enough time and data, can find good solutions. Demonstrating this in yet another experiment in yet another domain represents marginal value.
  • Skills in math and computer science. What subjects? Linear algebra, calculus, probability and statistics, algorithms & data structures, and a bit of algorithmic complexity. You should be good at solving problems and possibly at programming (if you are very good at math, I don't care if you can program or not; you can pick that up later). To check your problem solving abilities, in an interview I will likely ask you to solve simple math problems. Ones that I would expect a really good high school student to solve. If you can e.g. solve the logic puzzle, probability and elementary calculus and number theory problems that go under the name 'puzzle sheets' on this page, then you are in a good shape!
  • Knowledge of ML. This is required for PhD students and is a plus for Masters students. You want to prove that you have a good grasp of some of what is happening in machine learning these days. How do you do this? Pick a paper from a recent machine learning conference, like ICML or NIPS, or you can also pick a journal paper from MLJ or JMLR, read it, understand and interpret it and then explain to me what is in the paper. What are the strengths and the weaknesses of the paper. What are the remaining outstanding issues? Or why did you like, or you did not like the paper? Presumably the paper would be in your area of interest, matching some of my interests. ICML and NIPS papers are online, just like JMLR papers. E.g., you can find ICML'2007 papers here, NIPS'2007 papers here, and JMLR papers here. Luckily, these days anyone has a good chance to follow the literature because the vast majority of publications is freely accessible on-line.
  • Familiarity with some of my previous work. It is a plus, if you are familiar with my recent work. Look at my slides, my papers. How am I going to check this? You pick a topic and you will explain the contents to me. Again, I am interested in your ideas related to my work. So merely repeating what my work was about (or a few keywords) will not help you. My publications can be accessed here.
  • Good grades and letters, TOEFL/GRE scores I do care about grades and letters, at least to some extent. Same for TOEFL/GRE scores. Admission to our program is competitive, hence having good records on paper is insufficient. If you have bad grades in math or weak letters, you will not have much of a chance. However, I typically give chance to people who have some bad grades if they are likely to meet my other criteria (independent thinkers, skills in math, knowledge of ML).

What happens in a (phone/skype) interview?

I call you, we chat and I will ask questions related to the above items. You will also have a chance to ask questions of me. The purpose of the phone interview is to give me a chance to assess your knowledge, the way you think and your English skills. It is a good idea to be prepared to answer my questions related to the above. In the phone interview I will ask specifics. No need to talk about how much you love AI/RL/ML, ..., I know you do, unless you would not have applied.

Should you send me an e-mail?

If you think you meet the above criteria, I would love to hear from you, explaining why you think meet these criteria. Please discuss the criteria one by one in your e-mail. If we met before, you should mention that in your e-mail. In any ways, if you want to apply here to become a student of mine, you have to submit your application through the GAPS system. We do award fellowships to well qualified students for both our MSc and PhD programs. You can read about AICML, the machine learning center associated with the department here. Please note that it is not me, but the graduate comittee of our department, who is making decisions about admittance of students.

How to call me in your e-mail?

I know this causes a lot of headache. My preference is "Csaba". More information is here.

Advice for Graduate Students

A piece of advice by Michael Steele. An advice which is borderline harsh, but there is a lot of truth in it. An article about what research is like in the grad school. The take home message is that if people in the grad school are doing their job well, you will feel stupid. However, everyone in research feels stupid (not knowing what lies ahead), so grad school is for you only if you don't mind this feeling:)