Ask HN: Thoughts on grad school? (CS PhD)

To give a little bit of background about myself: Although I took some CS courses in college, I didn't seriously consider myself a programmer until just over a year ago, when I started writing code as a quant at a hedge fund and realized that, when using the right languages and working on cool problems, programming can be a lot of fun. I've also realized that there are many people a lot better at it than I am, so I have a lot to learn about technology and computers. Among my interests are artificial intelligence and programming languages, so I figure that graduate school might be the way to go.

I'm interested in AI/machine learning research, which practically require a PhD, even in industry. I've also considered the entrepreneurial route, but I lack a certain technical maturity as well as co-founders, both of which I could find in grad school. Eventually, when I'm in my 40s, I'd like to be involved in VC... but I'd like to spend the next couple decades doing more detailed technology work so that, when I get there, I understand what I'm analyzing.

So, given my interests and career aspirations, I'm considering pursuing a PhD in computer science. I'm wondering if anyone here could give me some perspective on what to expect.

Here are my credentials. I'd like to know what I can expect in terms of admissions success. Are, for example, top-10 departments a possibility?

3.9 / 4.0 in-major (math) from top liberal arts college.

CS courses: AI, PL, Software Eng., Data Structures, Intro. Grades of A/A- in all but SE.

GRE (2004): 800Q/670V, 98% on math subj. test, have not taken CS subj. test.

Top 100 finish on Putnam.

Research: Two internships in applied math, but in a govt. context. Consequently, I have no publications, so this may be a weakness.

Career: Entered pure-math grad program in fall 2005. Passed quals but did poorly in courses. Left after 1 year (no degree) to work on Wall Street, starting 5/2006. Worked on WS up to 2008, with a reasonably successful career and 1+ year of programming experience, but no research or publications (obviously).

Specific questions are:

1. How much is it going to hurt me that I entered a PhD program and left? Is this a deal-breaker, or can I explain it away? (My reason for leaving pure math is that I realized the academic job market sucked, and that I was already qualified for a decent industry job. However, a CS PhD carries a lot of benefits outside of academia.) Could I possibly turn my grad school experience into an asset, noting (in spite of my lack of success, and departure) that I've been through the experience before and know what I'm getting into?

2. Ideally, I'd like to get an NSF or equivalent fellowship, but I have no hope of getting that without research experience. Is it likely, outside of the ivory tower, that I can hook up with a mentor and get to work on a problem so that, when it becomes time (12/08-2/09) to apply to schools and for fellowships, I'm on the ball?

3. How much does the distinction between top (5, 10, 20) schools and the rest matter in CS, assuming that I will not be seeking academia? Do AI research groups care about the distinction between #1 and #8 and #22? If I pursue the entrepreneurial route, are higher-ranked schools going to have better co-founders?

4. The most important, but most distant question: What can I expect once I am in graduate school in computer science? What are the courses, students, research opportunities, and career possibilities like?

Have you read this?

"Applying to Ph.D. Programs in Computer Science", by Mor Harchol-Balter ( http://www.cs.cmu.edu/~harchol/gradschooltalk.pdf )

It's quite good. Since you have been in grad school already, many of the things won't be news to you.

I would say that the fact that you left grad school for a Wall Street job will hurt your application. The admissions committee might think that you're not determined and self-motivated enough for grad school. They might think "If this dude likes CS so much, why didn't he jump to the CS department back then?". Might be a good idea to explain in your statement of purpose that you discovered your passion for CS while in Wall Street.

Which is the longshot, finding the mentor or the NSF?

My experience is that top departments, and fellowships like NSF, basically want people who are already grad students. Actually, NSF seems to have more to do with the merits of the research problem than the student, per se. I don't think I have a shot without enough research experience to have a rough idea of what "my problem" will be.

2. Especially since you have a pure math background, you could work on SAGE (http://sagemath.org). The people there (mostly academics) are very friendly and open, and you should be able to find something with an AI flavour (e.g., automated empirical optimization of software, as is done in ATLAS http://math-atlas.sourceforge.net/) that has a fair chance of generating a publication, or at least generating an outcome you can point to.

I strongly agree. The PhD is an apprenticeship, so the people you'd be working for (and with) are the most significant factor. Make sure you fit with the research and social philosophies of a program before applying if you can, and certainly before accepting an offer.

As you leave school and prepare to enter industry or academia, my experience has generally been that the strength of the recommendations backing you matters far more than whatever rank your institution may have.

As you mentioned AI/machine learning, I believe these factors are even more important. There are some very distinct schools of thought when it comes to those things, so make sure you look for philosophical compatibility when picking programs. AI has a few deep schisms and widely separated sub-fields, so tread carefully.

It may be worth it to consider some of these: What's your philosophy of mind? Symbolic/statistical/neural? How important is biology/neuroscience when looking at artificial intelligence? What general approach to AI/machine learning most interests you?