A personal advice on how to be a successful PhD student in machine learning
Skills in programming and maths, as well as various other “hard” skills, are necessary but not sufficient to achieve success as a PhD student in machine learning. There are many other factors that also matter a lot, which I wish I understood better when I was beginning my PhD. As I am about to advise my first group of PhD students myself, I was wondering what advice I can share with them, which made me reflect on this topic personally. Here is my best shot.
Work hard. Get things done. It’s as simple as that. It doesn’t matter if you have lots of great ideas if you can’t turn these ideas into results and then papers. Getting research of top quality done requires dedication.
Take responsibility. You are the person who ultimately is responsible for your success or failure. Not your advisor, not your colleagues, not anyone else, you are.
Come to the office. PhD is a hard thing to do and it’s even harder if you don’t talk to other people in a similar position. It’s tempting to stay at home, but it’s usually not the right choice.
Don’t be shy to discuss your work. Speak to other people, professors, postdocs, other PhD students at your department about your work. They will often ask good questions that will make you rethink elements of your project. It’s dangerous not to periodically reflect on what you are doing.
Pay attention to details. It’s very hard to debug machine learning code. Don’t waste time on this. Do your very best to get it right the first time. It’s better to spend an extra day coding than to spend two weeks debugging or, even worse, failing to debug your code and, instead, concluding a good idea doesn’t work. The same goes for everything else that you do. Quality matters a lot.
Stay focused. It’s extremely hard to do top quality work on more than one project at a time. Resist the temptation to spread your attention over too many projects.
Be patient. It takes a lot of time and perseverance to successfully complete a research project. Very often this is not equivalent to actually answering a research question, which might require completing a few connected projects. It is a lot of work and you will finish only if you persevere.
Be modest. Keep it real. This is only the beginning of a beginning. Even if you are at the best university in the world, you are not entitled to a PhD yet. There were many people more brilliant than you who failed to get it. You will have to work very hard for a few years to earn it.
Be nice, make friends. Academia is a much smaller world than it appears from the outside and everybody knows everybody. You don’t want to have the reputation of someone unreasonable or rude. Just being polite and professional won’t get your papers accepted but it is important in the long term. People who you speak to at conferences will one day be your judges when you try to get a job or apply for a grant.
Don’t make foes. Unless you are very lucky, you will encounter an officemate that you don’t get along with, an admin person who doesn’t get things right or a computer support person who is not responsive. As a PhD student, you don’t have any power over anyone anyway, so getting angry and conflicting yourself with others won’t help anything. PhD on its own is stressful enough. Leave negotiating solutions to difficult problems to your advisor.
Be honest about what you don’t know. With yourself and your advisor. When you don’t understand part of a conversation, just tell your advisor. You are not expected to know everything from the beginning. However, after you’ve heard something a few times and didn’t ask for clarification, you are expected to know what the conversation is about.
Communicate clearly. The quality of your work is the most important, but even if it is the best research, if you can’t explain it, in a paper, in a conversation, or in a talk, your work is at risk of being ignored. Failure in science is never spectacular. Scientists fail when nobody cares about what they want to say.
Keep your advisor in the loop. There is a temptation to avoid speaking to your advisor until you finish a major part of your work. That might be a good idea if you know exactly what to do and you just need to execute it but definitely not if you are confused about what you should actually do. You advisor is there to help you. Even if you don’t need to talk to them, let them know what you are doing. It will keep them more involved in your project.
Educate your advisor. Your advisor doesn’t know everything but they are almost certainly clever and eager to learn. Teach them what you’ve learnt. If you want to get good quality advice, you must you give them enough background to understand what you are doing in detail. That includes showing them results of your experiments in the form, which is easily comprehensible. Create tables and figures that they can easily interpret. You want to leverage their knowledge and experience by allowing them to focus on the content, not on trying to understand the notation, axes labels etc.
Respect your advisor’s time. Don’t be late to a meeting unless there is a good reason. Academics are often extremely busy people. It’s not ok to be 15 minutes late. If your advisor can come to meet you at 9 am, so can you. Don’t assume that your advisor will fix your errors in code, edit a paper or a poster if you didn’t dedicate enough of your own time to do it. Don’t extract results from log files during your meetings. Use your advisor’s time wisely in a way that benefits your research.
Question your advisor. Advising is hard. Make sure that you have their attention and that they did think through your project. Ask them hard questions. They are used to that, they can take it and they will be happy to see that you are thinking deeply about your research. As much as you can you should try to be a partner in a discussion with your advisor.
Enjoy your life. It’s possible (but not advisable) to work 16 hours a day for a couple of weeks but not for a couple of years. Try to work efficiently for 8 hours a day and create time for doing things and spending time with the people you love outside of your studies. You won’t produce your best work if you are tired or miserable. By sticking to a schedule, your health and wellbeing will be better off in the long-term.
Good luck and enjoy your PhD adventure!
Acknowledgements. I would like to thank my own academic advisors: Amos Storkey, Charles Sutton, Rich Caruana and Kyunghyun Cho.