Defenses and Sensibility: Chronicles of a Thesis Supervisor
Episode 1 — Writing the Thesis Topic.
At the beginning of any thesis, there’s a topic. So, how do you write this darn topic? Are there any pitfalls?
Should you write alone or with help?
Consider if you feel capable of doing it alone or if your goal is to work with someone you know as a thesis director or co-director. In the latter case, it is crucial to build your topic with this person, after ensuring that you can share common goals. This will be the subject of another post! It’s important to be very clear from the beginning about how you see your respective roles.
When working with a funder, such as a company, certain points of vigilance are required. For example, the type of data to be processed or the type of results expected. Be careful if the doctoral student’s work becomes incompatible with the objectives of the funding group. It is necessary to consider that these objectives may also evolve during the thesis.
Anecdote: Georges had to work with a well-known large industrial group, which financed a significant data collection campaign. Georges couldn’t work with these data immediately because others wanted to first capitalize on them. In the meantime, he started working on another modality. The results came quickly. However, the company did not appreciate the valorization of this other modality, which conflicted with their own choices.
If you already have a good candidate at hand: involve them in writing the topic! You have everything to gain. In the past, doctoral students used to build their own topics. It was part of the adventure. It seems to me that this is still the case in medicine or law. I find this approach to be the most interesting and educational. Moreover, it allows to tailor the work to the strengths of the candidate and to reinforce their motivation
Starting well… it’s the starting point, not the destination.
True research does not know the destination, only the starting point. Therefore, choose a field in which you are well versed in the state of the art. Ensure that the topic hasn’t already been extensively covered in a work that answers most of your questions. List the outstanding questions you cannot answer about the chosen topic.
Research institutions and universities often ask for a very detailed thesis topic. It is not uncommon to be asked to plan your thesis over the next three years. Personally, I find this absurd for several reasons: a thesis involves a million unforeseen events, both personal and contextual. We know the starting point, but absolutely not the destination. Furthermore, some fields evolve very quickly, and the topic evolves with the candidate. For the same topic, you will get as many different results as there are different candidates.
So, why do they ask for this? To reassure themselves, or the funders.
You can pretend, write a generic plan with a state of the art in the first year, developments in the second year, and writing of articles and manuscripts in the third.
However, I believe this is a good opportunity to reflect on how the positioning of the thesis can be affected.
Encourage creativity and desire.
The best topics I have seen were the vaguest or the most daring. Take my PhD. topic for example: ‘convergence between polarimetry and interferometry’. Two techniques were new. They are mixed, we shake: what can it give?
Another idea: draw inspiration from a related field to advance your own. That’s what I did by proposing to ‘change scales’ to study polarimetry.
Crossing disciplines: sociology and data processing, art and science, physics and biology, etc. Science has no boundaries, interdisciplinarity is rare and will be increasingly sought after. So please yourself. Take something you love outside your research field and think about whether you can cross it with your discipline.
I have colleagues who had a blast doing experiments by riding bicycles or running through the woods in a panther costume. All of this for the advancement of Computer Vision. Yes, really.
Think about plan Bs. And if it fails, don’t panic.
In the past, care had to be taken to ensure that a candidate did not end up without data, or only with failed experiments: the nightmare of the doctoral student in experimental sciences. Today, in deep learning, there is a fear of posing too broad a problem that will be solved even before you start.
If, despite everything, this happens to you: do not panic.
I had a doctoral student, let’s call her Alice. Her topic was about improving image simulation using segmentation algorithms. At the beginning of her thesis, we were just starting with deep learning tools. We had our work cut out for us, just to adapt it to our data. In the meantime, SAM (Segment Anything Model) came out… and kind of killed the game! Other image generation tools blew up the scene.
You must have the courage to change course without throwing away the previous work and calmly explain this history in the context of this problem. There is always a way to bounce back. For that, it is more comfortable to address a particular type of data, or to consider an original use a posteriori of this feature, so as not to consider it as the ultimate goal. Huge progress has been made in your topic? Don’t get discouraged. Learn to use other people’s huge networks for an original use. This requires sometimes taking a step aside and taking a deep breath before continuing. Also, tell yourself that you are not the only one this happens to.
Note: These posts reflect only the personal viewpoint of the author and are his sole responsibility. The names of the participants in the anecdotes have been changed.
