My vision and practice of PhD advising

Mathieu Aubry , Aug 2023

Disclaimer: A PhD highly depends on the environment, the discipline, the people involved, the time…The following are personal views, about the way I see a PhD, and the way I typically interact with my students. This is not only subjective but very particular to my environment (Computer Vision in 2023 in a very privileged research environment)

I do not believe there is a universal goal to a PhD, especially not a quantifiable one. Every student is different, comes with their own personal history, view of research skills and goals, and will evolve differently. Each field also has specificities. The one I work in, Computer Vision (CV), has had an exponential growth approximately for the last 10 years, which comes with very specific issues and rhythm.

As an advisor, there are a few research-related things I want all my students to learn:

• how academic research, in our society and field, is working. This might seem obvious, but it is often quite far from people's preconceptions and I think the longer you practice it and you think about it, the less obvious, the strangest, and the more interesting it becomes.
• how to write, read and evaluate an academic paper, what are the necessary, dangerous or valuable things in them. There are a lot of implicit norms and skills that are necessary to write a (good) paper, but also to be able to read one meaningfully and efficiently.
• some expertise on at least a specific topic, and experience on how to gain expertise on a topic and be able to understand a specific field of research.
• how to build a research project, design experiments, analyze and validate results, see, select and push research questions.
• a personal view of what is valuable in CV research, beyond quantitative measures such as number of publications or citations, and beyond a specific publication.
• how to collaborate with others, and more generally how to interact in a research environment, including how to communicate about research and make presentations for diverse audiences.
• critical thinking about all of the above.
• probably quite a few other things I forget.

Of course, I also want them to contribute high quality research, papers, and code, and I want their research skills (including things not listed above, like coding and theoretical skills) to grow.

However, I think that beyond these points, a PhD is for most students a personal journey which I see as more important and difficult than the specific research output (though not completely independant). A PhD is a long time, with quite a fuzzy goal (publish? become a good researcher? graduate? get a job?...). It is also quite a solitary experience. Even if students collaborate on a specific project, they typically have very different PhDs and views. In many projects, a single PhD student is actually doing all the concrete coding and experiment work, with all the pressure and responsibility associated, and little possibilities for anybody to actually help or even understand all the details. Solitary and long of course means ideal for doubts and introspection, about projects, PhD, research, but also life and personal choices and goals. I think the possibility of having those is actually one of the great luxury of a PhD and makes it an invaluable experience, but it is also potentially very difficult to deal with. This is particularly true in the very competitive environment we do research in, and when everyone always tends to compare their own achievements to the most successful and hence visible people, which makes dealing with stress and sentiment of not being good enough key issues and an integral part of many PhDs.

How to teach the students all of the above points, help them produce quality research, grow personally, and survive their PhD, is supposed to be the role of the PhD advisor. However, how to do any of that is very unclear, and in practice the rest of the environment (other PhD students, lab, other collaborators...) has a key role. Thus, rather than tackling impossible questions about how a PhD actually works and what I try to do as a PhD advisor, the rest of the document simply describes concretely how my research group is organized and how I concretely interact with most students.

Concretely, my students are mixed with the ones (~20) of ~5 other permanent CV researchers with whom we share the same floor, which I will call 'the lab' (even if it's not officially a lab, I am skipping the 10 pages intro to the french educational system organization). At the lab level, we have a weekly 1h seminar with invited speakers and internal presentations. Students also have reading groups on specific topics, that are typically not followed by every student every time, and thus can be more irregular.

In addition, I have a weekly 1h group meeting with the students I am directly advising. These are the ones I typically count as 'my students/group', whether they are officially in their PhD or not (I have 1 to 2 Master intern/pre-PhD student every year that are integrated in the group in the same way as PhD students) and whether they are co-advised or not. There are many goals for this group meeting:

• for each student to realize the diversity of what the others are doing and how much time and effort it can take (e.g. when re-submitting for the 4th time a paper they did 2 years ago and that we struggle to get accepted, which is typically not something people brag about but is part of research), and to see diverse progression of diverse projects.
• for each student to get used to talking about their research (in public), and to explain it to people who do not closely follow what they are doing.
• for everybody to get feedback about their project, and learn to give feedback.
• for everybody to see what the others are doing, to potentially see opportunities for collaborations and synergies, to get ideas.
• to go more in depth about specific topics which are directly relevant to several people research.
• to discuss together group organization, updates and practical things.

I have about 6 students today and I currently think 5-6 students is my ideal size. I have had more than 10 and it was clearly too much, but I feel a minimum number is necessary to have some continuity, collaboration and group dynamic. Controlling the number of students is however much more challenging as it appears, since you might end up working a lot with some student without it being planned (e.g. if their advisor leaves), projects might have needs of a student at a specific time, some (few) student spend just 3 years at the lab, most around 4 (counting a 6 month internship at some point), some up to 5 years, I might have 2 great interns I want to hire and who want to continue in PhD when I was planning for only one, or none when I wanted to hire, and this is all hard to know in advance. I am also planning to have an additional postdoc and engineer for the coming years.

While each student typically has their own project at a specific time, there is a strong consistency in their research, which of course shifts with time but which I try to maintain. It doesn't mean there cannot be outlier projects or topics, research is a lot about 'chance' findings and unexpected ideas, but there are research questions and goals that I try to push on the group level. There are two main reasons for this. The first one is that I think this is very valuable for the students and the average quality of the group production, leading to many synergies, and this enables pushing and developing ideas beyond a single paper/PhD. The second is that I actually experienced students working in many different directions, and found that it significantly decreased my capacity to help them and give insights beyond the ones necessary to put a project in a publishable state.

I have a weekly 1h one-to-one meeting with most of my students (potentially with their co-advisor or current collaborators) most of the time. It would be hard to say exactly what the goal and content of this meeting is since it depends very much on the students, their advancement, and the content of their week's work. It could be chatting for 10 minutes about a stupid bug that's now fixed but they spent their week on, agreeing that the plan from the previous week still stands, and postponing the meeting. Or analyzing some results and looking at a bit of their code to be sure what it's doing. Or discussing long term goals. Or looking at a presentation. Or discussing reviews. Or many things. Most students either prepare slides or a list of points/results they want to discuss. For most students, we update a shared doc with plans/todo for the coming weeks and results/key insights from the work done. I find it helpful to keep track of a project in the long term, for me to quickly remember where we left things at the previous discussion, and prioritize the (typically many) things that would be great to do.

While every PhD is different and every student is different and changes with time, here is a sort of idealized PhD journey that I think works quite well for most students to learn as much as possible and limit the hardship and stress of a PhD.

A PhD usually starts with a topic, usually the same as the Master internship or pre-PhD the student did just before. However, in most cases this will only be the topic of the first project of the PhD. Indeed, CV is a fast evolving field, and in my area it is normal to work on projects that are typically at least 6 months and up to 2 years long (with start and end difficult to define since they tend to overlap with other projects) and each essentially corresponds to a publication (in CV we mainly publish 8 pages conference papers, associated to code, you can look at my website for examples)

Almost always, the topic of the PhD and the first project corresponds to a topic I specifically advertised and hired for, is reasonably well defined, and not very high risk. Even if they are still real research projects, this can sound a bit boring when you are a starting researcher and want to do a project that changes the world, but I think starting with such a project has many advantages:

• it enables me and the student to learn to know and trust each other, know and adjust each other's expectations, and learn to work well together. All of that without specific difficulties related to the project itself, which is good because it is often not as obvious as it sounds, and it is good to check we work well together before signing up for a PhD (since this starts usually in pre-PhD or internship)
• it builds up the student intuition of what is necessary to turn an exploratory project into a publication, and how to build up a scientific contribution. This can seem trivial but I neither think it is, nor that it is easy to learn in theory or by reading other papers.
• it can/should lead to a first publication, and thus a quick first hand knowledge of a full project-submission-presentation cycle.
• having a first completed project gives confidence and reassurance about one capability when being stuck later on a more open, complex and longer project.
• it enables me to help the student if there are issues and things don't work as expected (which might be much much harder on a more ambitious project) and know what are the things each student is good with or I should be careful about in further projects (should I allocate a lot of time to iterate on paper writing? does the student say clearly when they are not sure they have understood something? is the student overconfident in the absence of bug in their code? are there good practices I should double check they have? should I be careful to identify priorities and discuss time allocation for different things, or on the contrary do they work better when deciding for themself? ...)
• very often the project actually turns out not quite as trivial as expected and leads to interesting new research directions that can be directly incorporated into the project or be the basis for the next one.
• if it turns out the project really is easy and not that interesting, it can be done in 3-6 months, not much time lost and confidence and experience gained.

Further projects have much more variability, can be in the direct continuity of the first, come from a very different idea or have a very different goal. I think having difficulty, risk and student independence in the choice that increases progressively with each project is typically good because:

• the time of french PhD is very limited (going over 3 years requires asking for an extension, going over 4 years is strongly discouraged), so going too fast on a very open topic is very risky (even if one ideally already has a first project finished).
• there are many things that go into making a research project successful: while you can try to directly run, usually it makes sense to first learn to walk and progressively gain confidence, less risks to fall badly.
• the harder, the more open, the furthest from my interest a project is, the less I can help a student. The point of having an advisor is for them to be able to help.
• students progressively get better at and more confident in picking their research topics and less influenced by trends (which can be strong and I personally don't like to follow).

Ideally at the end of their PhD my students are better than I am at deciding what is interesting to do - at least on their specific research topic - and barely need my help, so they can go quite far from what I know/think/like. Also, failure on a specific project is not an issue since other projects are already finished. Of course, very few PhDs actually follow such a clear progression, and there are usually many opportunities and surprises.

People see and practice their roles as advisors in very different ways. I have seen people code their students projects, write their papers and almost their dissertations, I will not do any of that. I have also seen people not caring about their students, I try to.

Simply said, I think that the role of a PhD advisor is, precisely, to be an advisor. I will tell students what I think they should do (I might be wrong), but I will not do it for them, and ultimately every student is the one responsible for their own projects. This includes writing: I will provide feedback on papers, I might rewrite some parts when I think the paper is advanced enough, but out of very specific circumstances not before it’s nearly there.

As students grow into their PhD, I am happy that they listen less and do what they think is best. Growing into an independant researcher with their own idea is the goal of a PhD, learning also comes with doing ones own mistakes, and great project have happened that way. However, the point of having an advisor is to learn from them, so listening before arguing is probably a good strategy to start with.

There are quite a lot of more practical questions that this document doesn't cover (compute, conference travel, internal and external collaborations, policy on public release...). If you are actually joining the lab, we have an internal wiki for the lab and I have a doc for my students with much more practical information, feel free to ask for access.