Lessons learned from a PhD in Machine Learning
Part 3: How to get done
This is Part 3 in the series. You can find Part 1 here and Part 2 here.
Finishing your PhD
After a lot of research and hopefully some papers and conferences, it will eventually be time to finish your PhD. This can be really stressful, but it doesn’t have to be. To ensure smooth progress towards the finish line, you should start early (maybe even a year or so before your defense) to draft an outline for your thesis and discuss it with your advisor. If you study at an institution where you already have a committee at that point, you should also discuss it with them. This will ensure that you’re all on the same page regarding what should be in your thesis or not.
Once your advisor (and possibly committee) agree on a rough outline of your thesis, you are practically almost done. Of course, there is still a lot of work to do, but from this point on, it is very unlikely that you will not get your PhD. After all, your committee will not let you fail if you stick to the approved outline. While drafting your thesis, depending on the rules at your institution, it might be allowed for you to send draft chapters to your advisor and/or committee. This can provide you with helpful feedback and relieve your examiners from having to read your whole thesis in one go.
In general, when it comes to reading the thesis, you should also accept that most of your examiners will not read it cover to cover. So in the end, you might be the only person to have actually read your whole thesis at the point of your graduation. And that is okay. Think of it more as an exercise for yourself, where you can reminisce on the past few years of your research and weave them into a nice coherent story. Note that this will often not be the same story that you had imagined for your PhD when you started, but that is also okay. Note also that you have much more freedom regarding the content than what you would have in a research paper. For instance, would you like to include some weird little experiment that you did apropos-of-nothing in your second year? Or a funny alternative way you found to derive some textbook result? Or a comment about some troubles you had reproducing results from the literature? Then go for it! It’s your thesis, so you’re the boss.
One thing you can consider to make your thesis-writing more meaningful is to structure the introduction chapter as a review paper and make it independently available (e.g. on the arXiv). After all, by the end of your PhD, you should ideally be an expert on your narrow research area and you probably know more about the related literature than most other people. Since this is a very rare privilege, that you will likely not achieve in any other area in your life, you might as well use it to summarize the area for newcomers. Think of it as a stepping stone you leave behind for the next generation of PhD students. I personally structured my thesis introduction as a review paper, and have received a lot of positive feedback for it.
Finally, once you’ve submitted your thesis and hopefully passed your defense, try to reach out to your committee members for an informal debriefing a little bit later. They may still have some useful advice to share with you and can possibly help you with your next career steps. And after that, take a long nice vacation. You deserve it!
Next steps
After you’ve finished your PhD, you will have to decide what to do next. Many guides have been written about this decision, so I don’t want to go into too much detail here, but the main options are usually to stay in academia or go into the industry.
If you like teaching and supervising students (see Part 2), this is typically hard to do in the industry, and academia might be better suited to you. Also, if you have strong research interests that you want to work on, and you don’t want anyone else to dictate your research topics, academia typically offers that freedom. However, it should be said that faculty positions in academia are still hard to get (though in machine learning a bit easier than in other fields at the moment) and that you might end up doing several postdocs in different places and still will have to go into the industry in the end, because you just don’t get a faculty position. You will also usually need to be geographically more flexible and of course, the salary won’t be as high as an industrial one.
In industry, most positions will require you to do research on rather applied problems (since that’s how companies usually make money), but there are also a few positions where you can do quite basic research, similar to academia. Those positions are usually harder to get, though. Moreover, you won’t have the job security of a tenured academic, and your job description and set of tasks can often change over time due to causes outside of your control. Generally, if you like working on applied research (typically as a so-called “individual contributor”), you don’t like students, or you want to make a lot of money, the industry might be a better place for you.
As mentioned regarding the PhD choices above, a good strategy is to try out working in the industry (and different companies) through internships and to talk to a lot of people in different roles (ideally people with a similar background to yours) to learn how they made this decision and how they feel about it in retrospect. Also remember that you are never married to any job, and you should feel free to leave and look for different options if you don’t like the first (or second, or third) position you try.
So that’s it. These are the insights I felt were worth sharing. In the end, this series of posts turned out a bit longer than I expected, but to paraphrase Blaise Pascal, I would have made it shorter if I had only had the time.
This was Part 3 of the article series. You find links to the other parts of this series at the top. If you want to get in touch with me, follow me on Twitter or use the contact form on my website.
