Petar Veličković

Research Scientist


Hello! You’ve reached this page because you’re interested in contacting me. Thank you for your kind interest! This page provides useful information on how to most effectively get in touch.

It is strongly recommended to read this page first, before contacting me!

How to get in touch?

In nearly all instances, it is optimal to contact me via email: petarv at deepmind dot com.

Please avoid contacting me via LinkedIn, Twitter, or other social networks. I do not find such networks easy for direct communication, and in many cases I check them very rarely. When I do see a message, I tend to reply by asking for an email anyway.

It used to be the case that I could carefully reply to every non-spam email I receive. However, unfortunately, in recent times it has become obvious that this is unsustainable with the load of commitments I am facing. I realised a lot of my time goes into:

And I would like to optimise both of these cases as much as possible. In the below sections I will outline some common reasons for why you might want to reach out to me, and the optimal action to take:

Before proceeding, let me be clear: I am still very eager about receiving email, and try my best to reply quickly. This is for a good reason: my most exciting collaborations started by email threads. If I have not replied to your email within a couple of days, and it follows the guidelines below, please send me a reminder – it will be a useful signal for me.

Collaboration requests

It’s great to hear you are potentially interested in collaborating! I am always happy to discuss possible interesting avenues of future research – however, I am only really able to take on a few collaborations per year, and as a result, they would normally need to be strongly aligned with my research interests.

When writing your email, it would be very helpful if you make clear:

It is likely useful to consult my publications page while drafting this email. Keep in mind that some of my top-cited work may not be directly related to my current research interests.

If I accept the collaboration, it is very unlikely that I will be able to contribute by writing or checking code. For the vast majority of my recent collaborations, I primarily participated in meetings, helped steer the project forward and promote it, and co-wrote papers.

Talk invitations

Thank you so much for your interest! As I often like to say, the value of research diminishes without being able to properly communicate it, and empower others to make use of it. Hence I readily welcome any and all relevant talk opportunities.

In most cases, I aim to accept virtual talk invitations, and am also often happy to make most reasonable arrangements to give talks in-person (which I’ve found significantly more valuable).

When writing your email, it would be very helpful if you make clear:

It might be useful to consult my communications page, to get a feel for the kinds of topics I’ve taught in the past.

PhD applications

First of all, I am honoured that you are considering me as a possible advisor for your PhD work! However, unfortunately, my current position at Cambridge does not allow me to hire PhD students. Therefore, all requests of this kind will unfortunately be declined.

Internship applications

It’s exciting that you’re interested in applying for a DeepMind internship – and even more so that you are potentially interested in working with me!

I am generally interested in hosting Research Scientist interns (capacity permitting!) to work with me on a topic aligned with my research interests. Note that, to be considered for such a position, you generally need to be already in the latter stages of a PhD (with no more than two years left). If you do not meet this criteria, there also exist other DeepMind internships (such as Research Engineering) that you might find very interesting – but the pipeline for Research Engineers is significantly different, and not hosted by Research Scientists.

Generally, the application window for these positions opens around September / October and is quite narrow – please follow DeepMind’s social networks (e.g. Twitter) to be notified when the window opens.

It is useful to let me know of your interest before applying by writing me an email. When writing this email, if I am not already familiar with your work, please follow the same guidelines as for collaboration requests. It’s important to note that the application process has several independent stages, only one of which would involve a potential interview with me. All of this information is accumulated together when making hiring decisions, and therefore I am generally not in direct control over which interns get hired.

Lastly, you may be interested in getting a referral for your application. I am generally happy to give referrals to those researchers whose work I’m personally familiar with. Please reach out via email if you are interested in this.

Supervision requests

Thank you for your kind request – I am usually quite excited about advisory opportunities, especially if they align with my interests.

If you are currently a Cambridge student, and you would be interested in working together for your Part II / Part III / MPhil project, please do reach out! Note that my email address is not operational anymore, and therefore you should always contact me using the address given above. It would be very useful if, in your email, you could briefly outline your background, your interests, and what would you ideally like to take away from a project with me.

For all other students: please follow the same guidelines as for collaboration requests when writing your email. Further, keep in mind that, should I accept your proposal, I will be unable to be an official advisor, and you would need to find someone (presumably at your University) to act as the main advisor.

Advice on conducting research

I’m always happy to give advice and provide discussion on how to conduct research – especially for early-stage researchers. This has always lead to very stimulating conversations.

In the first instance, I found that the material I already made publicly available – namely, this Twitter thread – turns out to be very useful as a starting point. In addition, Aleksa Gordić put together several wonderful blog posts with advice for newcomers into AI research and engineering. Please, start by consulting those resources.

If, after reading the above, you still find that you have questions unanswered, or you would like to discuss further, please feel free to send me an email. It would be very helpful if you include useful context in your email, such as your background, and what would you like to take away from this advice. When appropriate, I may suggest we catch-up via a video call rather than exchanging emails.

Note that, very often, giving useful and targeted advice is quite challenging. As a result, and depending on my availability, you may expect longer delays before getting an answer – but I do try to answer to all such emails in a reasonable time frame.

Lastly, and very importantly – all of the advice I give is inherently biased by my own temperament, and my own experiences; shortly put, it is what worked for me. What worked for me may very well not work for you – your mileage may vary!

Materials on GNNs, GDL…

It’s great to hear that you want to enter the dynamic field of graph representation learning and/or geometric deep learning! In my opinion, it’s a very ripe time for making fundamental dents, and getting in on the action.

I have written not one, but two Twitter threads that summarise a bucket list of materials I would recommend for getting started!

For the most part, these two threads are still pretty much well-aligned with what I would recommend to newcomers, also stratified by which mode of learning they would find most useful (e.g. talks, blogs, books, courses…).

If, after reading the above threads, you still find that your specific need is not met, or you would like to discuss further, please feel free to send me an email. It would be very helpful if you include useful context in your email, such as your background, and what would you like to do with GNNs and/or GDL.

Happy learning! :)