Scott Freitas

Optimizing Your Machine Learning PhD Applications for Success


For many aspiring ML PhD students the application process is shrouded in mystery, feeling like a high stakes interview where you have no idea how you are evaluated.

When I applied to PhD programs in 2017, I had a million questions. How do I select a school? What’s the best way to find an advisor? Who is looking at my application? What are they looking for? How do I increase my chances of getting in?

After reading this post, you’ll be armed with the knowledge to answer all of these questions and more. We dive deep into the ML PhD application process, covering topics like:

For another good perspective on this topic, check out “Machine Learning PhD Applications — Everything You Need to Know”.

__________________

Is an ML PhD Right for Me?

The 3-7 (average 5) year journey comprising a PhD can be filled with breathtaking highs and seemingly bottomless lows. There will be moments of clarity where your insights tread new ground, and, if lucky, greatly impact your field. But more often than not, the days will slog by and you’ll begin to question the meaning of life. There’s a great number of reasons to embark on a PhD, and even more reasons not too.

Great Reasons

Not so Great Reasons

But really…why shouldn’t I do a PhD for the money or the small amount of prestige it begets? The reality is, there are much better, faster, and easier ways of obtaining these goals. Don’t get me wrong, almost everyone doing a PhD wants these things…it’s just usually not the sole desire.

__________________

Finding the Ideal Advisor

Arguably the most crucial part of the PhD is finding the right advisor. Not only are they your boss, mentor, and guide…but your only shot at a timely graduation. Below is my list of essential things to consider when selecting an advisor.

Tenured vs un-tenured

As a rule of thumb, the more senior the professor, the more hands off their advising style. Depending on your needs, this can be really great (no micro-managing) or incredibly challenging (what do I work on now?). Think carefully about your current research skills, and how much mentorship you desire. Some professors will expect you to independently explore new research directions, while others will want you to solely work on their ideas. Ideally, you’ll want to find a happy medium whereas you become more senior you explore more of your own ideas.

Lab size

Imagine two scenarios—Professor X has 2 postdocs, 8 PhD students, 3 masters, and 4 undergrads; Professor Y has 4 PhD students, 1 master, and 2 undergrads. Based on this, which lab do you want to join? You might initially gravitate towards Professor X, as he is likely established and well-funded. However, you have to consider how much time they will have for you. In all likelihood, your mentorship will largely come from the post-docs and senior PhD students. On the other hand, Professor Y probably has a lot more time to dedicate to you.

1-on-1 meetings

You might think it goes without saying that you’ll meet with your advisor on a regular cadence (e.g., weekly, bi-weekly), however, that’s not always the case. When in doubt, ask the current PhD students what their set up is. They’ll be an invaluable source of information.

Weekly meetings & socials

Lab social dynamics are critical to your success, and weekly meetings play a small part in this. Check if the lab has regular team meetings or other opportunities to socialize with your colleagues about research (e.g., to get feedback, brainstorm new ideas).

Summer expectations

Some advisors expect you to stay in the lab over the summer, while others will encourage you to explore industry. Check out where the current PhD students are spending their summers, as this is a great indication of where you will be spending yours. If you get a chance, ask them how they got their internship—it’s a big plus if you hear their advisor connected them to a colleague of theirs :)

Research alignment

When investigating PhD programs, it’s critical to match with an advisor that works in a field you’re interested in. Otherwise, you run the risk of doing research you don’t care about, or working on topics your advisor can’t advise you on. I’m not sure which scenario is worse, so lets try to avoid both.

Project collaboration

Do you envision your PhD as independent isolated work, or as a collaborative team project? Whichever you want, it’s important to understand your advisors expectations. Some professors encourage project collaboration, while others want you to go it alone (likely to make it easier to assign credit for what you personally contributed towards your dissertation). My personal preference is a 50/50 mix of collaborative and independent work.

Work enviornment

Do not underestimate the importance of your work environment. You will be spending the next 3-7 years of your life here. Are you going to be relegated to a poorly lit dungeon with no windows, or do you have a window seat with a nice view? Are you expected to show up to the lab every day at 8am and leave at 5pm, or can you work from home whenever you want? While these things may seem small, their cumulative effect can have a massive impact on your mental health.

Research funding

When applying to PhD programs, you probably aren’t thinking about your potential professor’s grant funding. Let’s change that. Continuing our example from earlier, Professor X is incredibly well-funded, meaning you get to spend your semesters as a research assistant. This allows you the opportunity to focus your time exactly where it’s needed—your research. Need expensive servers to run the latest deep learning experiments? No problem, there are plenty of GPUs ready to go. Afraid of being tied to a specific research grant, limiting the scope of your work? Not an issue, there’s a plethora of grants funding your lab.

Conversely, Professor Y has modest grant funding, meaning you’ll likely be a teaching assistant most semesters. Not ideal, since this takes valuable time away from your research (increasing your time to graduate). While there’s likely some servers and GPUs to run experiments, you’ll be sharing them with everyone in your lab, or even worse, everyone in your department. If you’re really unlucky, you may find your PhD funding is tied to a specific grant, limiting the scope of research you can explore.

While extremely competitive, you can apply for independent funding in the form of external fellowships (e.g., NSF GRFP, DOE CSGF, DoD NDSEG), or internal fellowships from the department itself. Coming in with this funding makes you extremely desirable to potential advisors, since you’re essentially “free” (or at least reduced cost) to them. To put it in perspective, if a professor is awarded a 500k NSF grant over 3 years, half goes to the school and half to the professor. If it costs 60-80k a year to fund a single PhD student, they can fund 1 full-time PhD for its duration. Who knew PhD students could be so expensive on such small stipends!

Applying to and winning external PhD fellowship is a whole topic on its own. If there’s enough interest, I’ll write up my experiences and advice on the topic.

Graduation expectations

No one enters a PhD program thinking they’ll be there for 6 or 7 years, but it happens all the time. Even if you’re hyper-productive, match with a great advisor, and finish all your graduation requirements in 3 years. Your advisor may want you to stay and continue publishing. It’s best to discuss any “unofficial” graduation expectations (e.g., number of papers, minimum time spent) BEFORE you join the program.

__________________

Understanding the Application Process

When you begin applying, it’s easy to get overwhelmed by all the required steps. Let’s shed some light on the process and break it down step by step.

Selecting the right schools

The first task is to find schools with the best CS/CSE/ML programs for your area of research. You’ll want to apply to ones with at least a few professors in your intended research domain, since you don’t know if a professor is going to see your application, want to work with you, or even take new students that year. When in doubt, send a thoughtful email to a few professors that you’re interested in working with, and let them know that you have applied to the school. This gives you a chance to enquire if they are taking any new students, and can help you stand out from the stack of other applicants.

When I applied 4 years ago, I scoured the department pages of 35+ schools looking for professors conducting research in the areas of graph mining and deep learning. In the end, I applied to 10 programs (UW, USC, UCSB, MIT, CMU, ASU, UCSD, Stanford, Georgia Tech, and Northeastern), and got in to 5 (USC, UCSB, ASU, Georgia Tech, and Northeastern). I recommend applying to at least 10 places, with an upper bound of 15. Any more than that and your letter writers really have to like you :) The #1 predictor of acceptance to a school is hearing from one of the professors. In fact, the only offers I received were from schools where a professor expressed interest in working with me.

Traditional wisdom suggests you avoid doing a PhD at the same university as your undergraduate alma mater (less of an issue for masters, unless it’s the same as your undergraduate). This advice stems from tenure track positions in academia, where there is a strong hiring bias against candidates who have stayed at the same university. While industry positions don’t typically care about this distinction, they do care about your schools ranking—where they frequently recruit candidates from “top universities”. That’s not to say that you can’t have enormous success outside these universities, just that it requires a little more work, networking, and luck. If you don’t know where to start looking, you can browse the U.S. News AI Graduate School Rankings (not an endorsement) and explore different programs and faculty members.

There’s also a couple of edge cases that frequently get asked:

  1. Should I apply to schools with higher ranking, or better professors?
  2. There’s only 1 professor I want to work with at the school, but they are a rockstar, should I still apply?
  3. The professor’s research seems interesting, but will I be a good fit?

Unfortunately, there’s no clear-cut answer to these type of questions. My general advice is select colleges that maximize your opportunities (i.e., more professors in your field) and to select professors you believe you’ll enjoy working with. At the end of the day, all you can do is make an informed decision and then move on. Give yourself a set amount of time to decide, and then don’t look back.

Increasing your acceptance odds

Knowing what professors are looking for in ML PhD applicants gives you a huge leg up against the stiff competition. While you generally need to submit four things,

  1. Two page personal/research statement
  2. Three letters of recommendation
  3. GRE scores (terrible requirement, fortunately some schools do away with it)
  4. Transcript

all they really care about is whether you will be a successful researcher. Take that in for a second. All these requirements are just noisy features to predict this one thing. If you can convince them you’re going to be the next rockstar researcher, you’re going to receive more acceptance letters than you know what to do with.

How can I convince them I’m going to do great research?

From a professor’s perspective, the best indicator is prior research experience. The gold standard is a top conference/journal publication in your field of research. If you achieve this, you will be a HOT commodity. The next best thing is working with a well-known professor on research and obtaining a stellar letter of recommendation. Only ask a professor to write you a letter if you think they’ll give you a glowing review, anything else, and it’ll do more harm than good to your application. If in doubt, ask them if they think they can write you a strong letter. More often than not, they will let you know if they can’t (it saves them the trouble of writing a letter they don’t really want to write). If you have a really great relationship with your advisor, you can consider asking your them for suggestions on places to apply, and if they would be willing to introduce you to colleagues at other universities.

Unfortunately, if you didn’t get a chance to do research as an undergrad it will greatly limit the odds of getting into the most competitive programs and labs. So much so that I’d consider doing a masters first, just to get the extra research experience. It’s not all bad though, this can be a great opportunity to see if you really want to do research, and some schools may even let you transfer over your graduate coursework. Plus, you come out of this with a master’s degree!

If you didn’t get a chance to do research and still want to apply, you can still get into great programs…it just requires you to have a very strong application package in other areas (e.g., GRE scores, classes/grades, internships, university ranking).

Interviewing with Professors

If a professor is inclined to extend you an offer to join their lab, they’ll likely set up some time to chat. This is normally quite informal—just a video call with the professor to discuss your research interests, background, and a chance for both of you to ask some questions. That said, there are exceptions to this rule, so be prepared for some professors to test your technical expertise. At a minimum be prepared to extensively discuss the following topics:

  1. Research background. How did you get interested in research? How long have you been in it? Do you want to continue working in this area?

  2. Professors research interests. There’s a good chance they’ll ask if you read any of their papers. Come prepared having read 1 or 2 of their most recent works.

  3. Coursework. They may ask about your coursework and try to probe your familiarity with high-level concepts.

  4. Aspirations. They’ll likely want to know your motivation for doing a PhD, in the hopes of mitigating the chances that you quit after joining. That’s a huge loss for a professor, especially a newer one.

They’ll likely ask if you have some questions of your own. I’d suggest keeping these light and asking 2-3 research related questions. You’ll have another opportunity to ask tough questions after receiving an offer letter.

Reaching out. If there are a few professors you really want to work with but haven’t had a chance to talk with, consider sending an email saying you applied to the school and why you’re interested in conducting research with them. If you have any papers, awards, or research experience, this is the time to mention it. The trick is keeping the email polite, short, and concise as to why you’re their next star student. If you did undergraduate/masters research and are close with your advisor, you may be able to convince them to reach out on your behalf (especially if they know the other professor).

Rejection, Acceptance, and Fellowship Letters.

Once your applications are out and the interviews are done, all that’s left is to wait. While everyone’s mileage will vary, a general rule of thumb is to expect a lot of rejection letters, a few acceptances, and (if lucky) a couple of internal fellowship offers from the department. If you got to the interview stage, and didn’t bomb it, chances are you’ll receive an offer letter. No interview? Well, probably no offer letter.

If you receive a fellowship letter in addition to your acceptance letter, congrats, you are a highly desired PhD applicant! The award amount will vary from program to program (maybe even between candidates?), but generally consists of tuition and stipend assistance for a fixed number of years. The benefit to you (aside from putting it on your resume), is that you are less likely to have to spend time being a teaching assistant (assuming your advisor isn’t rolling in grant money), and that you have a little more freedom in setting your research agenda.

Program Visit Days

Congrats, you’re in! This is the easiest and most enjoyable part of the process (next to receiving acceptance letters). The tables have turned and the professors are the suitors, and you are the catch. You’ll get to enjoy visiting schools where they wine and dine you in the hopes of convincing you that they’re your best offer. These trips normally take the form of (1) group visit days, where they bring everyone out at once, or (2) individual visits where they fly each person out one at a time. The latter is less common, so we’ll focus on the former.

Visit days are where you spend 1-2 days learning about the program, touring the research facilities, and meeting with the professors and PhD students. Don’t worry about the cost, it should be completely free—covering your airfare, hotel, and meals. This is where you’ll decide which program is the best fit for you to join. While you’re visiting, pay close attention to four things:

  1. Graduation requirements. Listen carefully when they discuss the graduation requirements, as this has a huge impact on the first couple years of your PhD. Some schools will have light requirements—a handful of required courses, qualification exam, proposal, and dissertation defense. Others can be more demanding, requiring an immense number of courses, teaching assistant requirements (cheap grading labor under the guise of “learning how to teach”), and a grueling qualification exam designed to fail a percentage of the students. Check to see how deadly the qualification exam is by asking current PhD students what their experience has been.

  2. Advisor-advisee interactions. See how your potential advisor interacts with you, and the other PhD students. This is a great opportunity to see first-hand how your future relationship with them will look.

  3. Lab social dynamics. Your fellow PhD students are going to be your colleagues, mentors, and (hopefully) friends. Spend this time getting to know them, and if you can envision spending the next few years with them.

  4. Location, Location, Location. You’re going to spend the next few years here, it’s important to envision yourself living here happily (or at least not unhappily). Do you prefer living in a rural college town, or a big city? Whichever your preference, try and get a feel for where you would live, and how you think you’ll adapt to this new environment. This might not sound important, but living in a place that you don’t like will wear you down.

__________________

Aligning your Post Graduation Objectives

After graduating, you typically have two career paths to chose from—(1) stay in academia and take a tenure-track position at a university; or (2) join an industry research lab, or applied scientist team.

Tenure Track Faculty Position

While I didn’t explore this route, I’ll share my limited experience and understanding on the topic. There are a few determinants that decide if you have the option to go into academia. Yes, that’s right, the option. Tenure track positions at research universities are highly competitive, and there are far less of them than their industry counterparts. The general rule of thumb is that you should have—(1) a handful of first author papers at top conferences or journals in your field, (2) the ability to communicate your research agenda, (3) great letters of recommendation from well-known professors, and (4) coming from a prestigious lab/school that sees many of its PhD students become professors. The harsh reality of academia is that it’s rare to see a PhD student from a lower ranking school become a professor at a higher ranking one. If going into academia is important to you, check how many former PhD students from that lab or department have become professors.

Industry Research Scientist

This has become the de facto route for PhD students post graduation. Your ability to join an industry team depends on your publication record, prior internship experience, how well you interview, and who you know at the company. Check out where the recently PhD graduates took jobs. Odds are you’ll be presented with similar opportunities.