One year ago, I wrote a simple blog post: I quit my job to start a PhD, and I explained in this blog post that I quitted my job at DernierCri for a PhD in Luxembourg on Automated Document Processing.
I was really excited about this new adventure, especially because I really wanted to be an expert in the (large) domain of Artificial Intelligence (Machine Learning [ML] especially) and Natural Language Processing [NLP], since my Bachelor Degree.
One year ago, I imagined my PhD as a vast field of experimentation with the following objectives: learn, experiment, fail, succeed, start over.
Narrator: It was not 100% the case.
One year ago, I imagined my PhD as a giant sandbox where you can experiment and discuss with guru to finally find a/some solution(s) to the problem you are working on, and publish it/them.
Narrator: It was definitly not the case.
Today, I am really disappointed about my research project and Research, and I just stopped my PhD to go back at DernierCri.
I did not want to create a blog post about my bad experience only.
Actually, if you have the opportunity to do a PhD: go for it! I strongly encourage you to pursue this noble quest.
But, please, do not make the same mistakes I did. To help you avoiding them, I have attached some tips below.
Do not accept a random subject
As I told you before, I really wanted to be an expert in ML/NLP, and I wanted to make something related with ML/NLP.
To accept a subject just because it’s related to an interesting field/subfield is the first (and maybe the biggest) mistake I made about this PhD.
Abstract your work
My three-years PhD was in collaboration with a (big) external company in Luxembourg, which has a clear and defined objective, and this has clearly an impact in your researches.
So, ok, there are good things in working on a research project with a company, like: you have the data and the physical resources to train and deploy your learning model; and I really think that Universities have to create a link with companies if they still want to innovate.
But, a company has an other objective than the research field: they want something to test and to push in production, with competitive results (first meeting with the company was like: “We want at least 99% of good prediction, for this complex task. Good luck, you have three months”).
With those kinds of objectives, you must organize yourself and draw a plan including deliverables.. And by taking the setbacks into account, you realize six months later you did not work at all on your research project. Furthermore, you realize the work you’ve provided for the company perfectly fits the company’s needs but not for you research project goals.
That’s when you realize in the end that you’re not a PhD student anymore, merely a weird mix of a lost student and a software engineer with a lot of big constraints from both sides…
Check the research team expertise
The research team I worked with was really great, but I found very strange at the beginning I was the only PhD working on documents classification, and information extraction from real-world documents.
Other PhDs, and graduated PhDs, had dedicated subjects about computer security, and used basic and advanced ML models to solve software engineering problems. Compared to them, my subject was more related to a ready-to-push-in-production project.
Also, I found sometimes that it was difficult to discuss with them about their research. As a personal though, since the subjects are very competitive at the moment, I think my colleagues were not inclined to share their ideas, to have a chance to join the podium in various conferences or workshops.
In this situation, I was alone, with the promess that a graduated PhD student will be hired as soon as possible to help me in my research project.
Today, no graduated PhD has been hired to work on this project, and loneliness ultimately wins over my motivation.
Check if your thesis supervisors are available…
…, at least once a week to discuss about your progress in your research.
In my souvenirs, I’ve had less than 6 meetings with my supervisors in the space of a year, which is incredibly little.
I don’t blame my supervisors, because I know that they are under pressure to set up projects and publish continuously for their own academic careers. However, it is better for you to move in an environment that is more likely to help you than the other way around.
Other researchers might don’t care about you and your work, they care only about your publications
As a PhD student, you will be evaluated by your supervisors (do not forget that you are a student). This evaluation will not be directly related to your code, the publication of your famous and innovative technical project, but to the papers you wrote and published (and where you published) during your PhD.
Also, as I’ve seen everywhere else, do not be surprised to find people you don’t know as co-authors on your paper - they are just short of papers for their “annual evaluation”…
It’s not because it’s a hard problem that is necessarily a research problem
Do not forget, just before applying for a PhD subject, to crawl Google Scholar or arXiv and to read scientific papers on. This simple tip can give you an overview of the innovation degree of the subject, and if people are close to solve the problem.
Maybe I wasn’t brave enough to support the pressure of both University and the company. But, today, I am feeling very good - physically and mentally.