Needing program advice; machine learning, computational neuroscience

[Sane]

Hey all. Hoping that I found the right place to post this. I was just looking for some smart people to get a few pieces of advice from, for planning out graduate studies. I finish my undergraduate at the University of Waterloo in a year's time. I'm currently planning to work my way to a PhD, but am unsure of where and what path I should take.

My background is in Computer Science and Combinatorics & Optimization. I find myself very interested in the field of machine learning, but feel it's important to also learn psychology and neuroscience. This is because of the many real life problems where the machine learning model involves a component of human interaction. I'll try to be specific: I have found the field of machine learning to be lacking in many of the problem spaces that are prevalent today, namely those where the underlying function that we are trying to learn is a product of the human mind (or even many human minds together). Here, it's important to take into account many behavioral factors in modelling the hypothesis space, and to keep it robust enough to allow for varieties of learning biases (there's no free lunch, etc, etc). Sadly, very little robustness is seen in most techniques, and the human component is generally viewed as more of a black box with algebraic or Bayesian learning biases.

I feel it's important to consider techniques that allow for more application specific modelling of the human component. I want to pursue a program that can allow me to explore some of these concepts in greater depth. I am unsure of what programs offer such a perspective. This is where computational neuroscience may come into play. But, I am not entirely sure, since I am only familiar with the machine learning side of things.

Ideally, I would be researching new approaches of modelling the human mind and applying those models towards specific real world problems, such as predicting user relevancy and preferences. These are problems that I find all around us, from Facebook, to News websites, to advertising, to our next-level media platforms (e.g. Netflix). These are problems that I know companies are asking, and machine learning literature has begun to answer (e.g. collaborative filtering).

Somewhere down the road, I wish to be doing research in these areas, as well as teaching Computer Science and Machine Learning as a professor. I'd like to know what school and program might be the best to get me there.

Hopefully that is enough rambling of where my mind is at. I'll just mention some of my credentials, to brag, and to let you know of my capabilities. My undergraduate is a 5 year bachelors co-op, double majoring in Computer Science and Combinatorics & Optimization. My cumulative average is 92% and my 3rd-4th year average is 95%. I have performed at an international level in Computer Science competitions, and have worked for 8 months for a startup in San Francisco as the lead machine learning developer. I should also be able to obtain a decent recommendation from at least one professor at Waterloo. However, I have no experience in Biology. I do have 3rd year experience in Psychology.

Sorry if that post was a little long. Just eager and looking forward to any thoughts. Thanks.

Edited March 24, 2012 by Sane

[intirb]

It seems to me like you have a couple of options. You could apply to CS PhD programs (it seems like you have a pretty strong background there already) with the intent of researching something sort of biological, or you could apply to Neuroscience PhD programs with the intent of researching something sort of computational. Both options are feasible, but for the latter you might want to take some courses or do some research that shows your interest in the biological aspect of the field.

My advice would be to look for professors that are doing work similar to what you want to do. If you can get in touch with them, talk to them about your ideas and ask them how you could make your application strong for a PhD program.

[Sane]

Thanks for the advice. Yeah, my background is definitely the strongest in CS. Would that generally open better doors than applying to something less related to my undergraduate experiences? Conversely, will having no experience in Biology hurt my opportunities in Computational Neuroscience?

This may also be a stupid question, but how much freedom do I generally get to pick courses and research? Does it depend on the supervisor? The worst case would be entering a Neuroscience PhD program, and not be able to (at least eventually) do research related to machine learning.

I know a professor I could probably ask about this, but unfortunately he's on sabbatical.

Edit: By the way, apparently Canadian PhD programs usually require a master's degree first, and also fund their masters students. Could I use that as my opportunity to get the breadth that I need?

Edited March 24, 2012 by Sane

[NeuroGal]

In many neuroscience programs there are people with physics, CS, psychology, and biology backgrounds. You would not be hurt at all to lack a biological background, especially in computational neuroscience. Most of the neuroscience programs will focus on the biology/psychology aspect and not CS which puts you in a great position. In fact, all of the people in my program who are in computational neuroscience never had psychology or biology backgrounds. They either came from CS, physics, or engineering.

Edited March 24, 2012 by NeuroGal

[intirb]

In general I agree, NeuroGal, but I've seen qualified CS people not do too well in their neuroscience PhD applications because the programs were worried that the students might not like neuroscience once they actually got a closer look. Doing something to show your commitment to the biological side of things (research, courses, whatever) would really help - just to prove that you know what you're getting into.

Every school and program is different. Whether you'd be better off in the CS department or the neuroscience department of a particular school, and how flexible your coursework and research direction would be, varies from school to school and even within a school between the different professors. If you already know what kind of courses you want to take and what kind of research you want to do, that's great! You can try to specifically look for programs that suit your needs.

Where do you want to do your PhD? A masters can be a wonderful way to strengthen your application (I got a masters before applying to PhD programs), if that's what you want to do.

[Sane]

It's a little hard for me to do anything now, because I'm on a tight track to completing my double major.

I will probably read some neuroscience books during my work terms. Maybe that will give me a better idea of what courses I need to take. Then, as you said, I could start looking for the right program.

I'm not sure where I want to do my PhD. Part of the problem. I could potentially do it anywhere. But Waterloo, or other Ontario universities, may be preferable, to be with family.

Waterloo looks like it has a really interesting neuroscience program: http://ctn.uwaterloo.ca/studies/gradprogram_details.html. It says it's an additional diploma on top of whatever masters or PhD program one might be taking. Ideally I'd be doing something like this, in conjunction with machine learning, throughout my master's and PhD.