Jump to content
Urch Forums

Advice/Chances in US CS PhD programmes?


Recommended Posts

Hi,

 

I'm thinking about applying to CS PhD programmes next fall. What are my chances for the top schools / can you mention schools that you think I could get into? (My dream is Stanford, but nobody gets in there...)

 

UG: Duke, B.S. Economics & B.A. Computer Science

Graduated with Honors & Higher Distinction for Econ Thesis

GPA 3.536 but 3.83 in CS courses

GRE: Q800, V720 A6.0

 

MSc Advanced Computing Research Stream, Imperial College London

(This is the #2 school in UK after Cambridge, well-known in CS)

Assume I get Distinction (edit - means all A on exams and research projects)

Courses include independent study, research seminar, etc.

Receiving selective departmental scholarship that covers tuition + stipend

 

Recs: 2 excellent recommendations from professors at IC. Can also get one from undergrad thesis advisor (gave me the Distinction)

 

Publications: none. This is my weakness. My master's thesis might result in a publication .. but can't assume that.

 

Area: Machine Learning, AI, Computational Neuroscience

 

 

So ... what are my chances? Also - I'm thinking about taking both Math and CS subject GRE in October. Is this a good idea? I've heard math helps..

Link to comment
Share on other sites

Regarding top ten schools, I don't know how master's theses are interpreted relative to undergrad students with publications. In any case, I would think you would have GA Tech in the bag already, especially if you are a U.S. citizen or permanent resident.

 

I'm thinking about taking both Math and CS subject GRE in October. Is this a good idea? I've heard math helps..
Technically, it won't hurt to take one of the Subject Tests since, at a minimum, you are just spending a Saturday morning and you don't have to select schools for automatic score reporting.

 

However, as you may agree, if you get a paper accepted by January, that would be enough time for the admin committees to consider it. So since you already are excelling in the GPA department, I would try to put every ounce of energy you have to spare into getting a paper published instead of stuyding the GRECS, especially as the GRE Subject Tests are a crapshoot -- it's very difficult to predict how well you will do, e.g. an extreme case:

 

I scored in the 41st percentile.

 

But last week I received acceptance into UIUC for a computer science PhD. With financial assistance.

 

...I have close to 20 papers, my undergrad was from a US State University not known for great research, my undergrad GPA was above 3.9, my master's GPA (from the same college) was above a 3.9.

 

http://www.www.urch.com/forums/gre-compu...s-not-end.html (Bad Subject Test Scores are not the End) (Bad Subject Test Scores are not the End)

BTW, you can only take one subject test for each test date, so, if you wanted to, you could take one in October and the other in November. However, it's pretty much impossible to study very well for both.

 

As far as math vs. computer science, I don't know. Even this hint from UIUC seems ambiguous:

 

GRE Subject. Applicants to the Ph.D. program are strongly encouraged to submit a score for the GRE (Subject/Advanced) Computer Science Test. If the background of the applicant is in mathematics, science, or engineering, they may submit an advanced GRE test score from that area.

 

Degree Admissions | Online Degree and Certificate Programs | Computer Science | University of Illinois at Urbana-Champaign

If you feel comfortable with the practice tests of both subject areas, I would suggest asking some of the CS departments you are interested in.

 

BTW, an old profile from 2004 for MS (not PhD) admission that makes me agree with you about Stanford (as their admit rate for MS is much higher than PhD):

 

I applied to 10 schools, heard responses from first week of February up until 14 April, and since all schools got back to me by the 15th, I was able to respond and go where I wanted.

 

Accepting me: University of Illinois - Urbana / Champaign, Brown University, University of Michigan - Ann Arbor, Columbia University, University of Colorado - Boulder.

 

Rejecting me: Stanford University, University of California - Berkeley, University of Washington - Seattle, University of Texas - Austin, University of Maryland - College Park.

 

No school offered me assistantships or fellowships, so my decision was based on my opinion of the school more or less. My undergrad was at UIUC, and I knew the program well, it was a good program. I visited at Brown and I'd seen Colorado in the past a few times. I chose to stay at UIUC, and being such a high ranked school, I know it can't have been the wrong decision.

 

My stats: 780/87% quantitative, 440/42% verbal, 810/80% CS subject, 3.33/4 cumulative, 3.5/4 technical, 3.75/4 CS only.

 

(original post from TestMagic no longer available)

Link to comment
Share on other sites

Thanks for your reply and your time - it's amazing that you can help out so many people here! BTW I couldn't make any of those links to other posts work .. they all gave 404 errors. I'll try searching...

 

Can we do something hypothetical? Best case scenario:

 

There's a chance I'll have a minor conference paper this year. So let's pretend that happens. Let's also say I take one of the two subject GREs (probably Math) and do well. Say my Master's thesis is quite impressive (like wins an award or something). Finally, say I have a great S.O.P. backed up by previous experience (I know this is optimistic, but never mind.)

 

Chances at:

- MIT / Stanford / Berkeley / CMU (I'm guessing still zero)

- UCSD ? UCSC?

- UofToronto?

- Harvard?

- UChicago?

- Columbia?

- UCLA?

 

Any other schools you can suggest that are urban, and preferably on a coast?

 

Cheers! (btw yes I'm a U.S. citizen)

Link to comment
Share on other sites

- MIT / Stanford / Berkeley / CMU (I'm guessing still zero)

- UCSD ? UCSC?

- UofToronto?

- Harvard?

- UChicago?

- Columbia?

- UCLA?

My Google queries for CVs of U.S. CS PhD students who got an MSc at Imperial College aren't helping me find anything, so I really don't know as I am less familiar with both PhD admissions and top programs. I did see this, but it was someone who went on to get a PhD in psychology:

http://www.wjh.harvard.edu/~kskassam/kskassam_cv.pdf

 

I also found a CV of someone admitted to the PhD program at MIT Media Lab, but that's not saying much regarding MIT for CS since the Media Lab is less selective and the person had prior publications:

Dimitris Vyzovitis

 

Since Imperial College ranks between Cambridge and Duke in the world academic rankings, I think looking at CVs of those who have done a masters thesis at a well-ranked university may help provide some clues (or just more questions).

 

All I can say at this point is the obvious, i.e. that you certainly would have a worthwhile chance at most places in the top 30. And, of course, in theory, there's nothing wrong with playing the lottery for CMU/Berkeley/MIT/Stanford if you don't let that decrease the number of applications you would be sending out to other universities, including other top departments like UT Austin and UIUC that have less competition compared to the top 4. In other words:

 

Apply to as many good graduate programs in your [research] area as you can. When in doubt, apply.

 

Advice for Undergraduates Considering Graduate School

Regarding anything below the top 50 like UCSC, with your profile, I wouldn't settle for it. Sure, UCSC has a great location, but I haven't seen any hot shot CS professors with a PhD from there except those who work in Bioinformatics:

http://www.www.urch.com/forums/computer-science-admissions/85858-ucsc-phd-computer-science.html

 

In any case, I would ask your recommending professors where their students have ended up. Even the ACM website says that doing so is a good idea since, as you know, computer science tends to be like Disney World ("it's a small world after all"). Similarly, I would guess your professors may have established good connections to some well-ranked U.S. CS departments, and, as such, your professors may have a better idea of some CS departments that may give more priority towards the type of AI research done at your university or the more specialized conferences you may publish to. In other words:

 

the schools that were great matches for my background and interests showed dramatically more interest in me.

 

http://www.www.urch.com/forums/computer-science-admissions/89511-lessons-learned.html

Of course, I am sure as you complete your thesis, you will become even more aware of which CS departments are most related to the research you are currently doing. Regarding machine learning:

 

http://www.www.urch.com/forums/computer-science-admissions/76926-can-anyone-suggest-me-some-good-universities-machine-learning.html

 

BTW, some rankings for AI:

Artificial Intelligence - Computer Science - Best Graduate Schools - Education - US News and World Report

http://www.www.urch.com/forums/admissions-results/46781-usnews-cs-systems-ai-ranking-needed.html

Ratings of CS graduate programs

Link to comment
Share on other sites

Thanks so much, you're really thorough. I'm going to look up that guy's Master's thesis, to see what I'm up against.

 

My gig is really Comp. Neuro, it's the coolest subject to hit the block since .. molecular computing? (I'd say quantum but that's kind of a dud). However, that's more Neuroscience than CS, yet for some reason nobody actually runs a PhD program in comp. neuro, you have to get into their CS dept first and then go work in the "MinDlAB" or whatever goofy 21st century name they have for theorizing about neurons.

 

Yeah I haven't been able to find much from Imperial either. I've also been unable to find CVs of people with PhDs from Imperial who are doing postdocs at the Dream 4.

 

I've got a serious dilemma now..

Link to comment
Share on other sites

  • 2 months later...

So here's my list right now, in approx. order of difficulty to be admitted, tell me what you think. The criteria for these places were:

 

a) There is a non-negligible probability of being admitted.

 

b) They have both excellent Machine Learning and excellent Theoretical Neuroscience research, with strong interdisciplinary collaboration (because I want to work on the edge of these fields).

 

c) In an urban center or at least a town that seems reasonable, with reasonable weather. I want a great lifestyle while I study my PhD!

 

Legend: (®r)r/m/s = ((really) ridiculous) reach/ mid/ safety

rr - U. Toronto

r - U. Illinois - UC

r - U. Texas - Austin

r - U. California - San Diego

r - Brown U.

m - U. Chicago

m - U. Washington - Seattle (fails ©)

m - Boston U.

 

Maybe:

rrr - CMU's PNC (Program Neural Computation) or Machine Learning Dept. (fails any semblance of (a)).

rr - NYU Courant (fails (a))

rr - Columbia (fails (a))

m - U. Waterloo (fails ©)

m - UCSB (fails (b))

 

I have a problem in that these are the places I really want to go - on the other hand, I'm probably not qualified to get into them (I've been looking at a lot of PhD students' CVs recently, and it's looking quite difficult ... even for the "m" schools).

 

So - I'm thinking maybe working in a research lab for a while, and possibly taking some math classes to boost my profile? Another option is get a Master's degree in Mathematics somewhere.

 

I want to get into the right school, rather than going for a school I'm not utterly psyched about.

Link to comment
Share on other sites

(I've been looking at a lot of PhD students' CVs recently, and it's looking quite difficult ... even for the "m" schools).
Yes, that certainly is true for University of Washington.

 

So - I'm thinking maybe working in a research lab for a while, and possibly taking some math classes to boost my profile? Another option is get a Master's degree in Mathematics somewhere.
Yes, since you haven't published anything yet other than working on a thesis, I would certainly try to publish something at a good ACM/IEEE conference, even a poster paper. So, at this point, I would be just concerned about getting published rather than where you should apply.

 

Therefore, the research lab internship sounds like a good idea, though I don't know how one applies for such things. Since the best indicator of research potential is getting published, taking more math courses would not boost your profile much compared to getting published and/or getting a strong recommendation from a well-known research lab.

 

In other words, a person with an MS in mathematics is not necessarily a good researcher. Though such mathematical sophistication can obviously help, actual research experience is more important than potential. Since for top schools there are number of applicants who got published during their undergrad, you want to at least be able to compete with some of their publication experience.

Link to comment
Share on other sites

Story of my life :) But, to at least make myself feel better about it, sometimes I combine the two. So if I am looking at a paper by a PhD student at UT Austin, I may peek at the student's CV. Or, vice versa, I will find relevant publications when scoping out department websites.

 

BTW, you may have noticed the upward trend in people writing papers or doing projects where they speedup existing algorithms by restructuring them for use on the CUDA platform for NVidia GPUs, e.g.

 

Evolved Machines uses GPUs for Simulation of Neural Computation :: GPGPU.org

http://www.evolvedmachines.com/images/news_32709.pdf

 

(I see that one of the alums from a UT Austin lab went to work for Evolved Machines.)

 

VideoLectures.Net: Fast Support Vector Machine Training and Classification on Graphics Processors

 

(There seems to be a lot of low-hanging fruit in the GPU general computing area. So I have bought one of the NVidia video cards for about $200 since GPU computation will probably be the focus of my first paper, and, in any case, I will use it for my genetic algorithms course.)

Link to comment
Share on other sites

Hi Meanlittlechap,

 

You mentioned earlier that you have a Thesis, I suggest you to start from there and write a Journal paper out of your thesis (specially you already have the problem, contribution and results) and this will take less effort than starting over from scratch searching for a problem.

 

Speaking of my experience, after I finished my thesis, I worked with my supervior on a Journal paper, it took around a month of deligent work until we formulated the paper and submitted it to some reputed Journal. Later when I received the reviewers report I realized this is what RESEARCH about. Journal publications reflect research maturity becuase the paper will be reviewed and criticized by two specialist in the field.

 

Having a research maturity is a requirement in order to crack top schools. Currenty you have the impressive profile, it's the time to work on research.

Link to comment
Share on other sites

Actually my supervisor is handling the communication with the Journal and as I haven't heard from him since we worked the reviewers report 5 weeks ago, I think it's still on the process.

 

Here are some related information:

Journal: Cluster Computing the Journal of Networks, Software Tools and Applications

First submitted: October 20, 2008

Received Reviewer's Report: March 13, 2009

Reviewer's Report submitted: March 30, 2009

 

See, the whole process would take at least 6 months.

For me, the most difficult part was the reviewer's report, they discuss evey single aspect of your paper.

Link to comment
Share on other sites

@CalmLogic

 

Yeah the GPGPU stuff seems really popular these days. A bunch of my friends are doing their Master's theses on it - e.g. source-to-source compilers and whatnot. One of my friends is trying to write SLAM using the iPhone's GPU, to do real-time augmented reality - I think that's a really cool project. That said, I'm not really a systems/HPC guy.

 

Also, honestly, I haven't investigated to what degree that sort of parallelism can easily be used in the stuff I do. I think it's quite specific to particular pieces of an app. So e.g. Monte Carlo or particle filtering are fairly easy to implement on a GPU with massive speedups, but for something like the EM algorithm you probably have to taylor the parallel computation to the specific model you're inferencing (the dependencies between different variables probably govern the amount of parallelism you can get ... that's just a hunch though).

 

There's a nice short Stanford paper by a bunch of students (undergrads? maybe) called "10 popular machine learning algorithms parallelized" or something like that - they recast all of them (including EM) into MapReduce, which is obviously perfect for GPUs.

 

What I'd looove to see is someone make a GPGPU compiler for data-parallel Haskell. Hell if I knew anything about languages and compilers I'd do it myself. Oh yeah, and Evolved Machines is awesome.

 

@Luma

 

You're absolutely right. I'm itching to get started on my thesis (right now I'm going through my final exams for this Master's degree), if only because I can work on something really cool instead of just taking classes (which admittedly are interesting). I saw Simon Peyton-Jones talk on How to Write a Paper, and was totally inspired by his idea - write the paper BEFORE you do the research. Then just do the work to back up what you said in the paper. So you can submit the paper as early as possible! How clever :tup:

 

I have some good ideas and if I work in a modular fashion, maybe in one or two months I can send something to a conference/workshop. Even so, if I submit something in e.g. August, it's unlikely to be accepted anywhere before the December PhD deadlines (actually, it's unlikely to be accepted anywhere at all :doh:).

Link to comment
Share on other sites

There's a nice short Stanford paper by a bunch of students (undergrads? maybe) called "10 popular machine learning algorithms parallelized" or something like that - they recast all of them (including EM) into MapReduce, which is obviously perfect for GPUs.

 

Cool :cool: I had seen that paper before, but I forgot all about it. (That's what happens when I don't take notes :D) Thanks for reminding me.

 

What I'd looove to see is someone make a GPGPU compiler for data-parallel Haskell. Hell if I knew anything about languages and compilers I'd do it myself.
Yeah, as you may have seen, there is a group in Australia working on something like that, but nothing is available for download yet:

 

Manuel M T Chakravarty - Data Parallel Haskell

Link to comment
Share on other sites

Well, maybe this approach would buy you some time, but be careful, this also could fire back on you. Editors in cheif nowadays are very picky due to the bulk of submitted papers. I trust you will do great in both of you exams and research :D
Link to comment
Share on other sites

DPH is working, btw ... the latest GHC compiler has an early version of it, and if you get the src tree you can get a latest version of DPH (substantially better than what's bundled in "stable" version).

 

It's extremely cool because it's so flexible. Here is parallel QuickSort in DPH:

 

qsortVect':: [: Double :] -> [: Double :]
{-# NOINLINE qsortVect' #-}
qsortVect' xs | lengthP xs I.              | otherwise =
 let p  = xs !: (lengthP xs `I.div` 2)
     ls = [:x | x       gs = [:x | x  p:]

     ss = mapP qsortVect' [:ls, gs:]
in
(ss !: 0) +:+ [:x | x 

 

Each of the list comprehensions like

ls = [:x | x 
 happens in a single parallelized step. So that line says get me all the elements of the array before than the partition.  But then, [code]mapP qsortVect' [:ls, gs:]
Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...