Questions from my ICML 2020 Mentorship Sessions

7 minute read

Published:

During ICML 2020, I offered two mentoring sessions over Zoom, which were attended by 20-25 people each. It was a great experience for me and I hope it was also great for attendees.

We also had asynchronous questions on Rocket Chat, here are some of the questions and my replies. I hope it is useful for more people. I have edited them a little in order to make sense for this format.

Hello Matias. My question is your advise for undergrads (from not too strong schools) trying to navigate their way through the field.

Hello! This is a very good question. I will make a summary of what I said in the Wednesday session about it. You should definitely try to get some basic knowledge in your field or sub-field of interest. I would start by reading survey papers, as they explain the basic concepts and they do a review of recent research, and I think this is the best way to get into a field. Unfortunately there could be no survey paper for a specific field, in that case you can try to find blog posts or online videos about the topic, so you get a complete understanding. Also important is to know that absorbing information always takes time, and you do not have to know or understand everything quickly. Give it time to understand and make connections between fields and ideas. This also related to Impostor Syndrome, as its easy to think you are not doing good when it is just a matter of time for you to understand a field and develop your own ideas.

Thank you, I have a question on a similar note i.e. if there was any advice you could give for someone that wants to pursue graduate school, but has no publications. Especially when there are no mentors or supervisors to discuss at school.

You should be able to pursue graduate level education (specially a Masters) without having any papers. But in order to do this, you should not aim for the top Universities and Institutions, where it is a kind of implicit requirement. Getting educated in other institutions is completely fine and could be a way to trampoline yourself into the field. By top I mean mostly in the US and Canada right now.

And of course, looking for online mentorship and collaboration opportunities should always be an option. For example connecting with people at this and other conferences. Many researchers have the problem of “having too many ideas” but not enough time/resources to try them, so sometimes just asking you might get to collaborate on a topic with someone.

How can someone without a supervisor/mentor go about publishing their first paper?

Writing a paper without support is quite difficult, but not impossible. You could leverage books about academic writing or grammar, or digital tools that improve your grammar. But also if you have a nice idea and some initial results, then you should be able to attract people to help you. Initially you should have ideas and skills to produce results, which also might be a problem. I know of cases of undergrads producing high quality papers to top conferences completely alone (ICLR 2017 was one case), but this requires a lot of motivation and hard work. Finding someone to collaborate and supervise would make things much easier. You could also try to get feedback online, for example in a Forum or in Reddit, but this has a large variance, you might get not so good or discouraging comments.

The good thing is that writing papers is a learnable skill and you get better with it over time. Your first paper might not be your best, but that does not matter, your best work is in your future!

Reviewing papers is also a related skill, and there have been workshops on this, where you can get experience on what reviewers look for, so you are prepared and write your paper considering this. For example https://sites.google.com/view/making-reviews-great-again/ And also this workshop: https://www.cc.gatech.edu/~parikh/citizenofcvpr/

Hi Matias! Do you think there are other ways/skills to show to compensate for not having papers when aiming to get into a great Ph.D. program? Even having done research, it can take a bit to have a paper published before the application deadline. I see you mentioned specifically US and Canada, is this “implicit requirement” not a requirement for top European institutions?

It varies a lot more in Europe than in North America, because the Education and Research Funding system work differently. Your first step in the way to a PhD is to do a Masters degree and for that you do not need to do it in the same institution, so there are more opportunities to do it in say, a middle tier institution, and then jump to a top institution for your PhD, specially if you do a good Master Thesis. With your Master thesis you have a document that can be used to compare you against other candidates, its not a paper but generally it has much work that can be published, so for PhD applications after a Masters we usually look at the candidate’s thesis.

And regarding the Funding System, since you do not go with a bachelors straight to PhD, and doing a PhD works in a different way than in North America, there are people that are working full-time in Industry and doing a PhD at a top institution (specially in Germany), and this is because their company has connections to the Institution (through a Professor that collaborates) or the candidate basically convinced the Professor that he can do worthy research. This is called External Candidates and since there are no tuition fees, it only depends on you and your ability to network and convince a Professor that you can do research, and generally its quite low risk for the Professor. This is mostly possible because of the Academia-Industry relationship in Europe, and the complete lack of tuition fees. In many German Unis (but not all of them) you can even do a PhD without registering as a student, just working with a Professor in a group or related company, until you have a thesis and submit it to the University to defend and graduate.

I also know that for PhD applications to many top Unis in Europe (like ETHZ/EPFL, Oxford, Cambridge, etc) it is much better to have previous publications, but these are easier to get while you do a Masters degree, instead of having to prove the same with only a Bachelor’s.

Overall, I think that if you have papers in the pipeline, even if they are under review, you could attach them to your application or mention them in your CV, so they are considered for evaluation by admission committees. The paper does not have to be published, it should at least be complete, as they can be evaluated by Professors which should be experts in the field where you want to do your PhD (and the paper should also be in their area).