Realism of Landing a PhD Offer

General Forum
1

Hi, everyone! I am a postgraduate at University College London, pursuing a Master’s in Machine Learning, and I will soon be applying for admission to PhD programs that start in Fall, 2025. I will share my profile and the schools I will be applying to, and am hoping to learn if the labs I am aiming for are beyond my reach.

I received my undergraduate degree in Mathematics and CS with first-class (honors) from Nanyang Technological University, Singapore, and am expected to earn my postgraduate degree with first-class (honors) as well. I am interested in theoretical deep learning – problems around curvature of loss surface, optimization trajectories, learning dynamics and generalization – which are mathematically intense research areas. Although my coursework has remained mostly theoretical and well aligned with such research (by design), my research experience has been more experimental. I have a third-author publication at ICML, on the work I did for my bachelor’s thesis project. It is a fairly theoretical work, but I was responsible only for the experiments. I also have a 2 first-author pre-prints – one experimental work on NLP (aiming for an IEEE publication), and another in graph ML (aiming for one of the top conferences), which has a decent theoretical component, but not as much as the work I hope to do in my PhD.

I am aiming for labs in ETH, UCL, Stanford, NYU, EPFL, Columbia and Princeton (in that order of preference, one of these is my pos). All of them have very successful PIs (by citations), who work on topics very well-aligned with my interests. My concern is that my seemingly all-over-the-place research background might turn them off, but I am hoping that my grades will convince them that I am competent with theory. I expect my supervisors to write excellent recommendation letters since they have appreciated me on numerous occasions. I am hoping to write a convincing research statement, but since I only started reading on relevant literature a couple of weeks back, it may not end up being excellent.

I don’t mind working with a younger PI, as long as I have some researchers working on adjacent topics around me. With senior labs, there is a network already established, and I can probably start by assisting on some projects, before getting into independent research. Realistically, am I punching about my weight? If I am, can someone suggest younger PIs working on aforementioned research topics, whose lab I might have a better shot of joining?

8 Likes
2

You’re approaching this from the wrong perspective. Instead of focusing on the program, think of it as applying to a particular research group. Without connections or direct contact with potential principal investigators (PIs), no matter how strong your profile is, your chances are still a gamble. Even if you get in, there’s no guarantee you’ll join the group you prefer.

A PhD is essentially a job, and how well you fit with a PI’s working style is crucial. Based on what you’ve written, it seems like you’re not looking to work independently right away. Some PIs, especially those who prefer close collaborators or assistants, might be okay with this. However, PIs who expect independent junior researchers may not find this a good match.

The only thing people on this forum can tell you is whether you’re clearly unqualified for a PhD, which doesn’t seem to be the case. Beyond that, the best way to assess your chances is by reaching out to the professors you’re interested in working with.

5 Likes
3

The field of machine learning is so saturated that: (a) most competitive applicants have at least one first-author paper at a top conference, and (b) every well-established lab receives numerous applications.

Having such a publication record doesn’t necessarily prove you’re significantly better than those without one, especially with the rise of large language models (LLMs) making it easier to get a paper accepted at a top conference. However, from a principal investigator’s (PI) perspective, publications often serve as a filtering tool since it’s impossible to conduct hundreds of interviews. So, if you don’t have lead author papers or strong letters of recommendation (LoRs) from recognized scholars, your chances are slim, even if you show strong potential. This is particularly true for labs at top schools like Stanford, where nearly every student has both credentials, sometimes even as undergraduates.

I suggest broadening your list of applications (seven is a small number; even five years ago, ten or more was common), unless you already have a backup. Also, you might be mistaken about the level of hands-on guidance in different labs: younger PIs can often provide more support because they have fewer students, are under pressure to earn tenure, and have only recently started their independent research careers. Senior PIs tend to run larger labs and are less likely to offer detailed guidance unless there is a clear structure for postdoc or senior-junior collaborations, which is often evident in their publications.

Don’t worry too much about the diversity of your research background—it’s common for students to explore different areas. If you’ve now found a clear focus, highlight that in your statement of purpose (SoP) and in your emails to potential PIs.

4 Likes
4

I’m at one of those schools and happy to give some advice about this, check pms/chats

3 Likes
5

You should definitely strive to get personal money for the programs you want to apply to. This makes you stand out from the crowd because, despite what you (MIT) say or the financing that your institution is receiving, I already have the go-ahead to turn this become a reality.

2 Likes
6

I have graduated from the same masters as yours and I have been a researcher at a top US ML university for some years. From my experience, in general, Europe and US operate differently in their PhD admissions. In the US you apply for a PhD programme (and not a specific topic) and need to get admitted by a committee. In Europe you apply to a specific PI and specific topic. In the US a PhD is around 6yrs and in Europe 4. A decade or more ago you could apply for a PhD in the US after your Bachelor but nowadays in top schools in ML is so competitive that students are pursuing a Masters of Research/Science with the intention to publish.

Studying in Europe, my guess is that you will have higher chances to get admitted by a UK university or other great universities in Europe. I know the Gatsby has a great programme and community along with Cambridge and ETH. Your main advantage is that people know each other and their research. Many of your supervisors or postdocs/TAs have strong connections with Google DeepMind which in the future my ensure you a great job as well. My experience in the UK also was that supervisors, even senior ones, pay attention to their PhDs and there will be weekly meetings. But because you are in the UK is better to talk with students or postdocs of them to get a feeling.

US has many advantages as well. Budgets aren’t even comparable to Europe. Getting admitted at a top institute will almost guarantee you a high paid job after your degree and lots of other opportunities. Some programmes at the beginning, allow the students to spend a few months at different labs to get expose to different research topics till you decide exactly ehat you want to do. You will also get paid a bit better than UK depending where you will live.

Don’t worry much about your “scattered research”. Math is a solid background and a great combo with the MSc in ML. PIs are aware that you are at your early beginnings.

My main recommendation is to choose carefully a place to grow and where your supervisor is expert in the field. Some PIs are looking to open up their portfolio hoping that with the student’s work they will publish in other sub-fields which have no background. It is very important for the PI to know the state of the art snd have a vision where it goes. The lab will consist of other researchers and if you don’t get along with your PI, the way they work, mentality, etc you won’t enjoy it.

1 Like
7

UCL is your greatest chance, thus you should focus on UK and EU schools. I’ve never seen a graduate of a UK university accepted into one of the top PhD programs in computer science in the US. Connections are important to US PIs.

1 Like
8

How many good research recs and what’d you rate them out of 10?

1 Like
9

In 2011, I too received my diploma from the same program! I would apply to universities in Europe for a PhD since it takes less time. As previously said, admission to US colleges is more challenging unless you have connections. Is M. Herbster still in charge of the MSc ML program there? His Evolutionary Systems course was fascinating, as I recall.