I switched to the wrong PhD program. What do I do now?
September 28, 2021 12:19 PM   Subscribe

Long story short (snowflakes below the fold &c &c), I left a CS PhD program at one university about a year and a half ago. This was partly so that I could live with my then-fiancee, and now my wife, and partly because I wanted to branch out. But after a year and change of trying to work with my new advisor, I'm realizing that changing programs was a big mistake. I'm not sure what to do.

So the more full narrative goes something like this. I started a PhD program in Computer Science at [midwest] in Fall of 2016 where I was to work with [advisor]. While there I wanted to work on fairness in machine learning, which I had discussed with him prior to starting the program, and he seemed on board with. However, since I didn't have a project fully fleshed out, he had me start on a different project on a topic to do with program analysis (something I previously knew very little about). Not precisely what I wanted, but he had a grant to work on it. But, I keep going with it, trying to follow his directions for about a year and a half. At this point, no publications have come out of it. At this point, he leaves to go into industry.

Enter [advisor]. I had taken a class or two with him, and generally liked him. As [advisor] is leaving, he tries to set me up with a couple different people who could be advisors at [midwest], and [advisor] is the one who sticks. I try to keep going with the program analysis project, but [advisor] doesn't see the potential. Instead, I propose a different idea to do with fair ML, which he is ok with. I do that, and defend my masters. The paper it's based on gets into a fairly competitive venue. Somewhere in here, my girlfriend applies and gets into a PhD program in a different non-STEM discipline, and moves out of [midwest].

At this point, I start thinking about moving programs. I like [advisor], but the thing I'm interested in isn't his main area. So I apply to other programs. This time one at the university where my girlfriend/fiancee works, [southern]. In my application, I write about the ways in which I see overlap between fair ML (broadly interpreted) and the area that [advisor] works on. I get in. [advisor] has offered for me to finish out my PhD with him at [midwest]. At this point, COVID has just started, and it's not clear what the future will look like. I decide that working with [advisor] on fair ML as it relates to his area of research and also living with my fiancee is the best thing for me to do. Fair ML isn't a thing that [advisor] really works on, but I think to myself "I already have a background in fair ML so that will be ok".

Now it's been a year and half since that's happened, and I've tried to make things work at [southern]. But I've found it difficult to work with [advisor]. For one thing, he's a lot more senior than [advisor] or [advisor], so he's a lot less available. Another, more fundamental, problem is that he doesn't really see what's interesting about fairness problems in ML. Basically, he wants me to say "this is what fairness means" and then work from there, whereas my interest comes in when exploring tensions between different meanings of fairness, and ways of deliberation about how to make things fair. This has meant that proposals I've had for projects I haven't really gotten any good feedback on or been able to pursue.

So I think I made the wrong choice. Apart from regret, frustration, and a certain amount of bitterness, I am not sure of what to do. I don't think that I can transfer again (one program change, you got unlucky, two is a pattern). Fair ML is an area I'm pretty attached to. I have capital-T Thoughts about it, but am at a point where I can't pursue any of them. I want a PhD, I want to do research, but the research that I've done for [advisor] has nothing to do with what I came here to work on. Most of the advice I've gotten has been of the form “keep your head down, do what you have to do to finish, and then you can do what you want” but that advice falls flat with me for two reasons: 1. I've been doing that, and it hasn't worked 2. It's not clear that I'll be able to do research after all this.

Any advice, either pragmatic career-type advice or advice about how to cope emotionally with the feelings of disappointment are welcome.
posted by phack to Education (17 answers total)
 
Basically, he wants me to say "this is what fairness means" and then work from there, whereas my interest comes in when exploring tensions between different meanings of fairness, and ways of deliberation about how to make things fair.

One way to approach your interest would be to frame projects as "IF fairness means X, THEN we get these results." That way you can explore these tensions over a long period of time by varying X across different projects. You only have to commit to a particular definition of fairness for the duration of the project.
posted by ewok_academy at 12:32 PM on September 28, 2021 [2 favorites]


As graduate school wears on, everyone gets increasingly frustrated and discouraged. I recommend you just try to finish your phd as quickly as possible with current advisor. You don't say anything about abuse, so your choice is better than a lot of potential options. Maybe just focus on finishing with a secondary aim of what your next career hop will be and making sure you have something lined up for the finish. Phds are long and frustrating for everyone I know, but it sounds like you're doing ok. A lot of people feel stuck a couple years before the end, and I recommend discussing more specifically what success looks like to your advisor, and doing that. Also, discuss with him, your committee and other profs in your areas of interest what your career path could be. Try to network with folks doing work related to options that interest you.
posted by Kalmya at 12:51 PM on September 28, 2021 [4 favorites]


Another alternative comes to mind for me. If you can find someone at a different university who is working in the area of fairness that you want to work in then propose a collaborative project with them. You might be able to gain both a collaborator and an informal advisor and/or external committee member while keeping your current advisor. This can also help with networking when it comes times to look for jobs in the area you want to work, as you will have contacts in that field.

You'd have to pitch it to your current advisor as "look, what an opportunity for us to have this expertise on hand!", and you'd need to have an idea for a collaborative project that the external person would want to work on.

I've also seen grad students take a summer and go work in another advisor's lab to get experience in an area the current advisor doesn't have; that might be a way to make contacts and develop the idea needed for collaboration above. If you can afford to pay your way for a summer, or if your current advisor would support this, it might be an approach.
posted by procrastination at 12:59 PM on September 28, 2021 [10 favorites]


I'm coming from a UK rather than US perspective, but I'd say that very few CS PhD students in my experience end up with exactly the project they originally proposed, and many end up on something quite different from their original idea - it's a natural result of becoming an expert researcher doing a long-term project in a fast-moving discipline, and changes in project or advisor don't reflect negatively on you at all.

Basically, he wants me to say "this is what fairness means" and then work from there, whereas my interest comes in when exploring tensions between different meanings of fairness, and ways of deliberation about how to make things fair.

I would read this as your advisor trying to steer you towards a practical thesis topic within the area you're interested in (the "pick one" phase of PhD supervision). You need a clearly-defined problem to solve, both in order to structure and write a thesis within a reasonable amount of time and to make it possible to examine with a reasonable expectation of success. ewok_academy's suggestion above is a good one; if that framing doesn't work, maybe there's some other well-defined and testable question you can pose within the area.
posted by offog at 1:23 PM on September 28, 2021 [4 favorites]


my interest comes in when exploring tensions between different meanings of fairness, and ways of deliberation about how to make things fair

This does not sound like original research for a PhD dissertation - it sounds like an extensive literature review. I can see why your advisor is concerned.

What qualifies as success or failure in your project proposal? What knowledge is advanced by completion of your project?
posted by saeculorum at 1:37 PM on September 28, 2021 [1 favorite]


one program change, you got unlucky, two is a pattern
Yes, but getting stuck with the wrong advisor will do more to torpedo your career than having to explain why you bounced around a bit more during your PhD. Especially now, and especially since you have several genuine Good Excuses for switching when you did. I'm not in CS, but is your career history significantly "flightier" than most of your cohort? Don't let fears of straying from the sanctified path fool you into the sunk costs fallacy.

That said, saeculorum's objections are worth considering. Would your current advisor be more interested in you developing a method for arriving at consensus on fairness that would be an input in the process of developing an ML algorithm or system? Or a method for reverse-engineering the model of fairness that a particular instance of ML is using?
posted by All hands bury the dead at 3:11 PM on September 28, 2021


This is a bad reason to try to switch advisors. It seems like you're having an actual academic disagreement with your advisor. This is what academic research is about: disagreements are good because progress can come from them. I disagreed with my advisor soooooo many times. My PhD students have disagreed with me sooooooo many times. Very often these disagreements led to interesting results. It seems like I was usually on the wrong side though...

It's your job now to prove that your approach is the better one; that it allows for building systems that address meaningful problems. Make an argument with your research and convince him. That is a completely reasonable job for a PhD student. That is, in fact, what PhD research should be about. It is perhaps the most important part of becoming an independent researcher.

It wouldn't be unusual in my experience for an advisor to push back against a student's proposed research directions as a pedagogical exercise. The entire point of the training is to make you an intellectual peer who can engage in these topics on an equal footing, and a great way to make that happen is to give you an opportunity to prove him wrong. I've done similar for my students (after getting to know them and carefully gauging their personality and emotional response to criticism!).

Note that I'm talking about disagreements about the academic discipline and the intellectual content of your work. You didn't talk about any personality conflicts or pathological behavior from your advisor; that's a different topic altogether.

The other thing to consider is whether he's spent any money on you. That could create some negative feelings in the long term, and potentially poison other people in the department about working with you. This is not as it should be, but it's something you have to consider.
posted by mr_roboto at 3:38 PM on September 28, 2021 [9 favorites]


It sounds like your research focus is too broad to get concrete enough research questions to pursue with CS methods. Narrow it down to something you can answer by doing experiments on public data sets, or formal proofs, or finding bounds, or whatever you know how to do and deem suitable. Start with toy examples. Bird by bird, in the hand, not in the bush, to mix my metaphors. Once you’ve formulated and answered a concrete question, however small, you’ll have a much better idea about what to do next. A more narrow question will also sharpen the focus of your reading.

And seconding the suggestion of looking for people to talk to and collaborate with, but again, start small, so you can be concrete instead of getting stuck in what sounds like a more philosophical approach that is interesting but not suitable for CS research if you can’t break it down.

Also, the advice of persisting you’ve gotten probably doesn’t mean you should continue as before, but that you should find different ways to make it through. Perhaps you can be the collaborator for another PhD student at the department, or find a paper and think about how you could improve it, or even give your advisor’s idea a go and see what you can do with it!
posted by meijusa at 3:45 PM on September 28, 2021 [4 favorites]


Academia these days runs on networks. Particularly in your dissertation, success (in terms of future jobs) is as much or more about properly triangulating yourself with respect to the right scholarly communities in the field, as it is about having Brilliant Thoughts about your subject.

You could have the most fascinating and original discoveries in the world, but if the average hiring committee doesn't quite know how to classify them, or they don't properly connect to existing research so nobody reads or cites them, or they do connect, but only to a slightly-outmoded, unsexy, or otherwise low-status corner of the field, or in ways that aren't perceived as moving the field forward in the way existing tenured profs want it to be moved... well then, you're going to have a tough time.

If your advisor is medium-senior, odds are good he understands the networks a lot better than you do. If he's extremely senior-- enough so to be a little out-of-touch-- then he's still likely to be swayed by good reasoning in terms of existing networks, i.e. you explaining the value of your desired research with reference to the other major research it builds on, the fairly popular (thus, proven feasible) methods it employs, the specific other people at other schools who would totally be excited to read this, etc. As a bonus, once you know who those people are, you can also reach out to them to get their thoughts. Then you'll start having a network of collaborators who can also give you sanity checks on your topic and throw interesting side projects your way, and it won't matter as much whether your advisor is perfect for you.
posted by Bardolph at 3:52 PM on September 28, 2021 [1 favorite]


And, from birds to goose=gander fairness, I’ll toss out some examples of starting points.

Browse through the programs of the recent FAcct (previously FAT, fairness, accountability, transparency) ACM and ML ACM conferences, see what the scope of the papers is, what methods they use, what questions come up, what you could emulate, what you could improve, what you could reproduce, what small part you could add.

Take one of your fairness notions (which are you considering? Could be anything from plain proportional to, i don’t know, formalizations of the veil of ignorance or epistemic justice, just pick one) and look at possible tradeoffs and value tensions to explore, maybe accuracy, or privacy, or explainability, or accountability, or transparency, or performance. How does it play out on different datasets, different types of data (categorical, numerical, mixed, images, dense, sparse …) different ML approaches (decision trees, neural networks, deep learning, statistical approaches …).

Look at the recent EU AI regulation proposal, how does the notion of risk relate to your notion of fairness, pick a concrete application from the unacceptable, high or low-risk list to explore.
posted by meijusa at 4:12 PM on September 28, 2021


I'm not in CS, so weigh everything I say with appropriate skepticism, but I switched grad schools and have formally mentored PhD students in the physical sciences going through rather similar things. (I don't claim to have mentored them well.)

Questions I'd ask if you were a student in my department are:

1 - If you don't already have an MS, can you stick it out long enough to get one? It may obvious to admissions committees (and future hiring people) that something weird happened, but leaving with a degree certainly helps make it less weird. It will count in your favor when admissions committees ask the question, "will this person be able to complete things?"

2 - Are there flexible solutions that will let you work on what you want without changing schools? If you have a strong relationship with someone at another school, consider arranging for a local advisor who will sign your forms while you're actually working with someone remotely. Maybe that's not your current advisor, but it might be. (This is obviously a whole lot easier if you have your own funding or are willing to TA the whole time. If you haven't done so, start applying for your own funding immediately.)

3 - Do you have close connections with people doing the work you want to do? In the very short term, that'll help you decide about applying to another school. In the longer term, it's vital. If you haven't been doing it, go to every relevant conference you can find. Apply for grants and student support, but doing it out of pocket isn't a bad choice if you're really set on this subfield. Get to know people who are doing the sort of work you want to do. They'll either help your application get considered if you change schools, or they'll help you find the next job after you graduate with a less-than-ideal project.

4 - Have you talked to other people in the department? Some chairs are assholes, but not most of them. And even the bad ones are really concerned with student outcomes. If you ask for a 15 minute appointment and say, "I'm really frustrated and thinking about switching to another school," there's a reasonable chance they'll work to find a solution you're happy with. If there's an academic or student affairs vice chair, that may be an even better starting place. (Usually they've volunteered for those jobs.)

Best wishes, and good luck. Finding out you've made a mistake in grad school is a terrifying experience. I've been there. But, at least in this case, there are no mistakes that can't be fixed.
posted by eotvos at 4:16 PM on September 28, 2021


(Sorry - re-reading the question, I see the MS is done. Ignore my first point, and consider the rest in light of my poor reading comprehension.)
posted by eotvos at 4:24 PM on September 28, 2021


You want your thesis work to be on a problem just interesting enough that you can stay motivated to finish it. No more interesting than that, because more interesting is harder, and it'll be hard enough already.

Can you write a summary paragraph for a hypothetical dissertation coming out of your work? What type of result are you describing, perhaps a proof about the implications of different definitions of fairness? Demonstrating the effectiveness of a particular deliberation process?

If you can write up your possible contribution, and your advisor is questioning whether they sound like CS thesis results... get advice from people active in the field, your advisor could be off base, but asking this question is very much their job. Can you show published results that are similar to what you might achieve?

If you can't be sure yet what you'd even find... that's not abnormal, but an advisor wants to see clear traction in your exploration.
posted by away for regrooving at 10:17 PM on September 28, 2021


You know, I think my core advice might be: do you have any idea who might be on your committee? Who you might co-author conference submissions with? Find those people, working in ML fairness, and engage with them.

First, for your path within the Ph.D. program. If your work is not in your advisor's core area, that can work, but then you want at least one committee member for which it's totally their deal. Convincing that member that This Is A CS Result is what lets everybody sign off on this work even though they don't totally get its position within the field.

Second, more importantly, for your research itself. It's hard to work effectively alone. It's lonely. It's hard to orient without knowledgeable feedback. I hope you can talk or work with other candidates in your area as well as that committee member!

Third, if you want to keep doing research in the area after your program, these colleagues are how it happens.
posted by away for regrooving at 10:36 PM on September 28, 2021


I started a STEM PhD but for a variety of reasons did not finish. One of the things I learned too late was that sometimes it's the type of practice and methodology that matters more than the specific topic.

I don't know enough about CS research to know if there are multiple significantly different methodologies of approaching thesis projects (I'm sure there are?), but what I wished I'd done is chosen an advisor who used the research methods I ultimately wanted to employ instead of the advisor who was trying to answer the exact scientific questions I was interested in. There is more flexibility to switch topics after you graduate by saying "[X] skills and [Y] approach can be applied to topic [Z]." But you have to get papers and you have to graduate in order to do that (if you want to go into academia or get an advanced ML position, anyway).

So I would say if you think you'll get the skills and methodologies you need from this advisor and you think you'll get publications demonstrating such then hang in there. You have your whole life to pursue the problems you're interested in but you do not have infinite chances at getting the PhD you might need to pursue them in the way you want to do it. Very few grad students get The Perfect Project. 90% of it is being able to gut it out through the shitty bits.
posted by Anonymous at 4:09 AM on September 29, 2021


Gonna give right-angle advice here: how badly do you want the PhD? Why not stop? Do you have any interest in working in industry? Machine learning is the very hottest topic there is at the moment and places are paying stupid amounts of money for trained academic experts. Particularly since you had a successful master's project with a publication.

You could very easily work on the problem you're interested at one of about ten top tech companies. There's advantages to doing this kind of work in a company; enormous amounts of real data, a chance your work will actually change a product people use. Lots of drawbacks too of course.

I dropped out of my PhD program at the MIT Media Lab and went into industry. Never once regretted it. Definitely was the right financial choice. But also the right choice for me. I was having trouble getting a PhD project to click, and finding good committee advisors for my work, and generally just not thriving. Turns out the real problem is I shouldn't have been in a PhD program at all.
posted by Nelson at 7:03 AM on September 29, 2021


"keep your head down, do what you have to do to finish, and then you can do what you want” is bad advice. I mean, it's good advice if you plan to leave academic research altogether and reinvent yourself in a new industry after your PhD, which is what many unhappy students too. But it doesn't sound like that's what you want to.

Find another professor or mentor to work with. You don't have to make a drastic switch. Can you do a summer internship or visit another university? Also, make friends with your peers at conferences and see if you can collaborate with them. If your advisor isn't very hands on, he may not particularly mind if you have a side project (and that side project may grow).
posted by redlines at 9:22 AM on September 30, 2021


« Older Identify mystery photo?   |   My 10 yr old got a credit card offer in the mail.... Newer »
This thread is closed to new comments.