Should I try again next year; or join a sub-optimal PhD program?

[frustrated_indian]

Re-posting here for better visibility. I'd appreciate any advice and comments

I am an EE student at a top IIT with interest in AI/ML/vision. I have a high GPA (9.2+ can be considered equivalent to 4.0 at most US schools), good research experience, and a journal paper under review. I was initially interested in communication systems and information theory, and found my calling for AI very late. Though communication, info theory, and AI might sound quite different, there is a vast overlap in the underlying math - especially optimization and probability theory. I went ahead and applied for 6-7 CS PhD programs (mostly top 10) and haven't heard back from any of them. Nearly all of them have sent out their admission offers, and hence I am forced to assume a reject.

I have heard positive news from Columbia and UT Austin EE programs, which I had chosen as backup options. But I was contacted by only communications professors, even though I had mentioned my intent in working with a few CS professors in my SOP. Should I take one of these offers and try to make something out of it? I no longer have a passion for communication, and would like to work on problems in vision, robotics, and such. Also, it would be very unfortunate to go the EE route to these universities, since many of my classmates with weaker profiles got offers from better places like UIUC, Caltech, and CMU.

An alternate option seems to be to work for a year and then apply again in CS. I have a neat job lined up at IBM, but it is not full time research, so may not help with applications. Are CS programs averse to taking non-CS undergrads? What would be a good option given my circumstances?

I feel so pathetic and miserable. If I had to go the EE route, I could have got much better programs. If only I had taken a few AI and ML classes sooner, I would have spent time there and got good papers, instead of wasting time with communications. I can say very proudly that I have worked much harder than many of my peers, and I am at least as smart as them. But ultimately, everything seems to boil down to dumb luck - being in the right classes at the right time, and dumb luck with choosing the right major at 19 y/o without any idea what it is about.

[pterosaur]

If the offers you get don't match what you want to do, don't go somewhere that you won't be happy with. It's not like undergrad where you take the best of the offers and it's uncommon to wait and apply again the next year.

You're right that your problem may be that you're applying to CS programs as a non-CS major, which means you really need to highlight your CS experience in your application or perhaps build up experience on that side. You've got a good job offer, so it's not like you'll be left in the gutter if you don't take up the PhD. It sounds like you'd be better off applying again next year with a better sense of what schools are looking for, more experience under your belt, and with a refined list of programs (perhaps some that are more friendly to non-CS undergrads). I think framing your experience and personal statement and talking to your recommenders can be a big help in how your application is perceived. If the issue is that you don't have the necessary pre-reqs for CS PhD programs, you might also consider taking courses as a non-degree student before applying again.

[parasolsherry]

Yes, applying for CS PhD with a EE-background is VERY difficult (much more difficult than applying for a CS ms), especially when you're targeting at top10. Now that AI, CV, and machine learning has become so famous, only a few CS students with very related and decent research experience would get in. I know a lot of student with accepted, first authored CVPR/ICCV papers (i.e. top conferences in CV) get rejected by a bunch of schools. So presumably convincing your POIs that your communication background proves your potential success in AI field will be unfortunately hard. 

I would say accepting phd offers in an area you no longer feel excited with might not be good. I started as a computer engineering student before transferring into CS, and as far as I know communication is considered far from AI, and it will be less likely for you to find perfect collaborators in this case. However you may try transferring into CS phd later. I know some of my friends did this, but it's highly risky since you are not sure if you could find a professor nice enough to take you in this way. 

If you are not working in the IBM research lab, I guess the job might not be as help as you would wish, but still this sounds like a better option. If you could work as a RA with some AI related professors for the next year (maybe try to talk with someone in your undergrad university), that would be best for reshaping your research background. Taking CS courses will also be very helpful as CS professors will always want people satisfying pre-req. If your ultimate goal is a PhD and you dont mind paying for next two years, you may consider make things up by applying for some more FALL 2016 CS MS program. I know Columbia, UPenn, Brown and some other schools are still accepting students. There are also programs accepting students in spring. 

I get that it must be very depressing now that your choices are not as shiny as your classmates' as you decide to change the field, but still finding what you like right now is not late!

Edited February 12, 2016 by parasolsherry

[frustrated_indian]

@parasolsherry Thank you so much for the detailed reply. I was really down, and your comment helped life me up.

My job at IBM is soft of research, but not full time research. I am in IBM smart planet group, where I'd be both developing software and analyzing smart grid data to find strategies for demand response, time series analysis for forecasting, and optimization methods for dynamic resource allocation. Some guys in this role have told me that it is possible to get a co-authorship in papers or patents, but they will not be in traditional machine learning venues. Popular venues are IEEE smart grids, IEEE CDC, SIAM (including SDM) and such. So, I am a bit worried if this will actually take me even further away from the type of AI problems I wish to study - vision, robotics, SLAM, and such.

Would it be a good idea to contact Columbia now, and tell them that I am interested in pursuing an MS in CS, while also TAing in the EE department for funding? I have an admit with Columbia EE (with RA/TA), so not sure how it would fly. I am not in a position to afford an unfunded MS. I don't have an obligation to earn, but I am in no position to take out a loan or ask my parents for money.

[parasolsherry]

48 minutes ago, frustrated_indian said:

@parasolsherry Thank you so much for the detailed reply. I was really down, and your comment helped life me up.

My job at IBM is soft of research, but not full time research. I am in IBM smart planet group, where I'd be both developing software and analyzing smart grid data to find strategies for demand response, time series analysis for forecasting, and optimization methods for dynamic resource allocation. Some guys in this role have told me that it is possible to get a co-authorship in papers or patents, but they will not be in traditional machine learning venues. Popular venues are IEEE smart grids, IEEE CDC, SIAM (including SDM) and such. So, I am a bit worried if this will actually take me even further away from the type of AI problems I wish to study - vision, robotics, SLAM, and such.

Would it be a good idea to contact Columbia now, and tell them that I am interested in pursuing an MS in CS, while also TAing in the EE department for funding? I have an admit with Columbia EE (with RA/TA), so not sure how it would fly. I am not in a position to afford an unfunded MS. I don't have an obligation to earn, but I am in no position to take out a loan or ask my parents for money.

My pleasure! =)

Your work will actually include pretty good research, but from what I see it's related to general machine learning or data mining in practical fields. If you've fixed your interest so specifically on vision and SLAM, You are right that it is not so related to what you want.

Yes I understand the necessity of taking only funded grad programs, and I must say I agree with it. For the Columbia thing, I guess you could try, but EE and CS should have separated admission office and direct transfer at this stage while maintaining TAship in EE might not work out. Directly applying for CS MS with funding again might even be easier XDD

Or you may try entering Columbia as a EE MS student, and then to take various CS courses and work with CS professors. While your degree is still in EE, as a MS student you are less committed to your program than a EE PhD, and thus could get more access to CV related things! This will also help with your recommendation letters. 

I struggled wandering around many different research fields and only the last one I tried during my whole undergrad seemed to work out. I also have difficult times when I thought I should be doing better than my peers given my efforts, while in reality nothing seemed to pay off. Things go right and wrong now and then, but there must be a way to fix it! I believe ultimately you will be rewarded (and probably outperform your peers who now hold better choices), and the time you spend on communication may also come back in a positive way sometime unexpectedly. Hold on till that moment!