What I learned from an AI master degree - Part 1 (Finding a Project and Supervisors; Creating the Proposal)
Part 1 : Starting Troubles - Finding a Project, Supervisors ; Creating a proposal (This article) Part 2 : Timelines - Optimal Planning and Breakdowns Part 3 : Setup and Tools - Setting up Writing, Programming, and research to prevent tears Part 4 : Writing and Programming Advice
Introduction
So you decided to pursue a master degree? It must be scary to know that you have to write some obscure research thesis on a niche topic that your friends will give you weird looks about. Welcome to the club. I was pretty terrified too, but it all worked out in the end, and I now have a master in Artificial Intelligence from the University of Groningen, Netherlands. (Yay for me!)
This is a series of articles where I share what I learned from the project and what I would do differently if I were to start again. This is no definitive guide, but I hope it helps you, my dear reader.
This article is Part 1 : How to find a project and supervisors, plan your idea, and pitch it without going crazy. Depending on when you read this, the other parts will be out, and the links will be updated at the start of the article.
Pick your adventure. I am proud of you for whatever you do.
Finding a Project
The first hurdle you will face is finding a project. Of course, there is no point in thinking about the rest of the steps if you don’t know what exactly you want to do. So, how do you find a project? You have approximately three options.
- You can find a project on your own by finding a problem you want to solve or an idea you want to explore.
- You can find supervisors who have projects that you find interesting and ask them if you can work on them.
- You can find a company hiring interns and working on a project they have as the thesis.
Each of them has their own advantages and drawbacks. For instance, if you are not particularly comfortable in the field you chose, it would be harder for you to judge what kind of projects are feasible for you. In that case, you might be better off finding an existing project. That way you will be able to learn faster and also have a supervisor that is already familiar with the field.
On the other hand, if you are particularly interested in a field, you can find a research gap that you want to explore. In that case, you might be better off finding a supervisor who is interested in the same field and pitching your idea to them. If they think it is too ambitious, they can help you refine it. (This is what I did.)
If you are more interested in the industry, you might be better off finding a company that is hiring interns. That might give you some work experience off the bat and might (depending on how you work and the connections you make) also help you get a job later on. Depending on where you are based, this might be a little harder to accomplish. For instance, if you are in a country where either the language or the culture is different, it might be harder for you to find a company that is willing to hire you. Or perhaps you might find one, but it might take you a lot longer than you expected. If you are a native of the country, that might not be an issue. But if you are an international student, perhaps it is nice to have a backup plan and contact a few supervisors as well.
But What Ideas Do They Want?
The whole point of a thesis is to solidify your foundations and help you find a field you would perhaps enjoy contributing your time and energy to. So, it is not necessary that you have to find a project that is completely new. Many universities don’t impose much of a guideline except for stating that you should use at least some of the concepts you learned in your master. There is no “wrong” project. There is only a matter of feasibility in the time that you have. You might have a great idea, but if you don’t have the time to implement it (even partially), it might help to save it for later. Sometimes you may have to change your idea to make it more feasible. But that is okay. (This is what I had to do, as my initial idea was too ambitious.)
It would be nice to find something that you are interested in. Perhaps talking to your seniors or faculties might help if you are not sure what you want to do.
Okay, so how do you find a project? You have the following options. Pick your adventure.
On Your Own
Okay, so you chose to find a project on your own. Yay! But how do you actually find something to work on? Well this depends on what kind of project you want to do, but I will list down a few things that might help you out.
- Finding a Literature Gap : Simply put, this would be to dig through research in your field and find problems that they were not able to solve. Finding such a research gap does take time and effort, though. In some sense, this is also what the supervisors do, except they are more familiar with the field and might have an easier time in understanding the papers. (*Look at the “Future work” section of papers.)
- Merge Projects : Suppose you find multiple projects that do similar/different things, and you want to see what happens if you combine them. That is a good start. (This is what I did.)
- Continue Existing Research : Another way to find a project is to look at existing research and see if you have any other ideas that you can add to it. They might not solve the problem at hand, but even finding solutions that don’t work is a good start and might lead to one that does.
- Company Research Projects : Many large companies publish a lot of their projects as blog posts. For instance, one of my favorite ones is Google AI Blog. I have been following them on and off for many years now, and I have found many interesting projects that I would like to work on. This route may be helpful for you as well.
- Seniors : Almost all universities have a thesis archive. You can look through them and see if you find something interesting. But of course, you can also ask your senior friends for inspiration. You may be able to continue their work or find supervisors they worked with.
Find a Supervisor
Okay, so you chose to find a supervisor. For many people, this step is one of the hardest things to do. (Depending, of course, on the institute you are at and how many faculties are available.) In my opinion, it would serve you to make some relationships with the faculties about as soon as you can. This could be as simple as trying to find out what kind of research each of your faculties does. At the end of the day, they are here for you, and if you show them some interest, they might be able to guide you more than you think.
In my case, I remember getting this piece of advice from a senior during my bachelor, and since then, I have put a little extra effort into getting to know some of my faculties. I have to admit this made it easier to find a supervisor. My mentor not only guided me throughout the thesis, he helped me find a second supervisor and also trusted me to do my own project. He taught one of my favorite courses, and I remember just asking him how I could learn a little more in it, and he invited me to his office for a much longer discussion. To be honest, I had no idea he would be my supervisor after a year. I was just interested in what he was teaching.
The Proposal
Almost every university requires you to submit a proposal before you can start your thesis. This is a document that outlines what you want to do, how you plan to do it, and what you expect to achieve. It is also a document that your supervisor will use to judge whether or not your project is feasible and if they can guide you through it. Once you have the initial idea, creating a proposal will not take you that long, but what might take time is being able to phrase your idea in a way that is understandable and creating supporting diagrams, etc., that would make it easier for your supervisor to understand exactly what you want to do.
Depending on your supervisor, getting an initial approval may be quick or may take a little bit of time. It would help to maintain communication throughout. You are probably not the only person under them, and they are genuinely busy people. While you wait, there is no harm in trying to flesh out the rest of your project. It might save you a little brainpower in the long run.
Some Tips for the Proposal
While there is no “correct” way to write a proposal, there are some things that might help you out. I will list down some of the things that I found useful.
- Look At Examples : Simple, but perhaps not always done. You are not the first person to do a thesis. There is no need to reinvent the wheel. Look at examples of previous proposals (ask your supervisor, perhaps), and save your brainpower for later. Trust me, you will need it.
- TMLIF : Aka, tell me like I am five. How would you explain your idea to a five-year-old? Start there and then make it to the standard of an adult. Assume that the person you will present it to only knows about the basics and builds up from there.
- Literature : It would be nice to have a small initial literature survey to show that your idea has some support from the research community. This means finding some papers that do similar things or have concepts that you will use. Since you will need to do this properly later on, just finding a few relevant papers is enough here.
- Timeline : In most cases, a timeline is requested. While this is personal and my approach will not work for you, I recommend that you take into account resting time. This was something I only encountered later on, and it would have helped to think of it before. A thesis is a long task. You will lose motivation many times. It is quite normal, but if you don’t plan for it, you might end up feeling miserable for no reason at all.
- Template : Most universities have a proposal template or guidelines you can follow. While they are not always strict, it is a good place to start. At least you will know what is expected of you.
- How Much Is Too Much? : If you are doing your own project, it is hard to know when to stop. Just remember that this is not your entire project. This is just a draft, an idea. As long as what you have conveys your idea and how you plan to tackle it properly, that is enough. Keep it simple and short, stupid. :)
An Example: My Thesis
While my thesis is by no means a standard. If you are interested in Computer Vision, or want to see an example, feel free to look at the links below. The project was about emulating Attention from Transformer networks in regular CNNs using gradient-based XAI techniques as a proxy - aka “Proxy Attention”. (Yeah, I tried to make it sound fancy. Come on, just let me sound cool sometimes. ;p) You can find the paper here and the code along with all the writing here.
Fin
This article is in the hopes that it will help someone out. Maybe have the help that I did not. I do not know who it will reach. But to whoever it does, best of luck :)
Like these/Want more? Buy me a coffee! Kofi
Want articles on something specific? Just ask!
You can always contact me on LinkedIn, or drop me an [[mailto:msubhaditya@gmail.com|Email]]