How to NeurIPS
Or any big ML conference for that matter. Learnings from a very flabbergasted overwhelmed first-timer
So, last year I attended the notorious NeurIPS conference. A conference with over 8000 attendees, and endless amounts of posters, talks, and workshops. Let’s not even get into the number of sponsors. You’re surrounded by the gods of AI research (for the crew I was attending with, it became a game to spot them), and you often, if you’re feeling brave, get the opportunity to talk to them. Besides your most well-known 10 or 20 “god-status” researchers, you’ll run into a whole array of people whose research you’ve read and cited — often only realising much later who you were talking to.
If I had to describe NeurIPs in one word, I’d say “overwhelming”. As a first-timer, I was just so excited to be there and I annoyingly fell into one giant trap: I failed to plan.
Don’t get me wrong, it’s not that I didn’t desperately try to plan. I seriously tried. I opened up schedules before the conference and tried to read titles. I sat on the plane going through talk abstracts and workshop descriptions. Everyday, I’d open up the schedules and scan desperately trying to find words I understood or peaked my interest. On a daily basis, I’d inevitably end up missing talks that were super-valuable, or lost in a sea of posters — too tired to find the one I really wanted to see; mindlessly wandering through the reams of sponsor booths, trying to find friends I hadn’t seen yet.
Some of my inability to plan was due to the fact my attendance was last minute. But most of it was sheer paralysis — paralysed by choice, paralysed by impostor syndrome, paralysed by the fact that it’s impossible to try grasp everything on the schedule.
In the next few days, I’m attending ACL. It’s not near the size of NeurIPS, but I wanted to make sure I planned properly and decided to reflect on how to do that.
As per my “How to analyse a research paper” and “How to solve a hard bug” posts, I had a “non-strategy” for tackling NeurIPS. This “non-strategy” for NeurIPS was totally useless. It was a sort of random walk, where given the talk I’m currently in, and the talk descriptions I can physically speed read about in the next 2 minutes (a local search, since there are simply too many), which next session seems most beneficial.
For the main conference tracks, this didn’t work out too badly since the invited talks stood alone, and the contributed talks only had 3 tracks. On the other hand, the workshops and the posters were something else.
I tried to see everything and instead exhausted myself and ended up seeing very little.
Additionally, after a long day of conferencing, anyone who has attended NeurIPS knows of the legendary (but also quite garish) sponsor parties. There are a couple each night and getting tickets to them is some black magic that involves chatting up recruiters, finding the event websites 3 weeks before the conference, or knowing someone from the inside (an article for another day). And the parties are amaaazing!
So that hopefully sets the scene of chaos. I spent the last 2 days in Montreal talking to other researchers and trying to learn about their strategy for tackling a conference of that magnitude.
#1 Plan Poster Sessions in Advance
I ran into the brilliant Dr Katja Hofmann and I asked her how she dealt with the conference and didn’t come out overwhelmed. She said she did significant preperation for the sessions. For every session, she’d have already read all the titles and some of their abstracts and picked maximum 3–4 that really seemed valuable to her, and she’d go spend her time there. She said that additionally, instead of going to every evening event, taking time out in the evening to plan the next day was very valuable.
#2 Take an evening (or two) off
A continuation of Katja’s advice, take some time off. Either to prep for the conference, or go have an early dinner with friends. The main conference is an assault on your mind and body, so getting enough sleep and resting is important
#3 Go to parties, but be Cinderella
At NeurIPS, the events are legendary. Ostentatious, and over-the-top, but eventually they made me giddy (and not just because I can’t say no to a good cocktail). I should have applied the same trick I did during university days: Leave the party before midnight. That way you’ll still get enough sleep to handle the coming days. Having little self-control and lots of excitement, I would get carried away and be left with the regret of being unable to concentrate during the morning session.
#4 Have conference goals
I ran into the excellent Dr Vukosi Marivate at an event and he told me he was meeting up with a sociologist and a political scientist so they can collaborate on a piece of work. Muthoni Wanyoike and Dr Benjamin Rosman were often meeting up with people to talk about the Deep Learning Indaba.
I eventually noticed a trend: People had deliberate goals beyond just listening and learning. Afterall, it is so rare to have such a gathering of minds caring about the same thing and I learnt that a conference is much more than the sessions and parties you attend.
Some goals that I came across:
- Community building & support: I attended both the WiML and Black in AI workshops. I spent time building connections, listening, and making plans for future engagement. It was fantastic to watch and support research by young and inspiring researchers in the field from Africa. It rejuvenated me. If you’re feeling overwhelmed, find your people: They will help centre you in the chaos.
- Find collaborators: What a better place to find people to work with!
- See the leaders of the fields talk: As someone from a country far far away, the opportunity to watch the leaders in your field argue amongst themselves on a panel are few and far between!
- Explore new and fascinating research areas: I ended up in the Causality workshop. I didn’t have half an idea what was going out, but what I did understand, I found fascinating. I bought “Book of Why” on the flight home.
- “Exploit” on your field: By “exploit”, I mean “exploitation vs exploration” — Find other people who get really excited by the same things as you and exchange struggles, and ideas, and ask detailed questions.
- Meet your idols: And make sure you have a selfie with them. Or ask them the question you’ve been dying to ask (or be like me and embarrass yourself by going blank as soon as Jeff Dean is in front of you)
- Find or become a mentor: There is nothing more valuable to a young researcher than having someone give them feedback on their work. If you have the privelege of expertise, a conference is an amazing place to mentor young researchers. I know, for example, in many institutions in Africa, students sometimes do not have supervisors (or at least supervisors with experience in their field), so some kind and constructive feedback from someone who knows the ropes is extremely valuable.
#5 Fight the Imposter Syndrome
As someone not coming from Stanford, from a far away part of the world, who’s returning to research from being in Industry, the sheer amount of intelligent experts around you becomes overwhelming at times. There are sooo many topics you’ve barely heard of and have no idea how they work, that there is no way you’ll ever be able to understand everything that is going on.
It’s easy to start telling yourself that you don’t deserve to be there and that those researcher’s would never think you have anything valuable to add. And there where definitely times where I’d wallow in this fact.
It was only when I wandered into the Emergent Communications workshop to watch Nando De Freitas talk, did I realise that my strange background in swarm intelligence (my MSc was around communication strategies of ants and bees) and newer experience in NLP gave me a unique perspective to the work being presented there.
You have value to add. You deserve to be there. Open yourself up, instead of hiding. No one knows everything. Ask stupid questions. Grow.
So, let’s hope ACL is a bit less overwhelming this time :)
Laura Jane Martinus and I will be presenting a poster of our work on Benchmarking Neural Machine Translation for Southern African languages at WiNLP
If you’re interested in how we managed to get to NeurIPS, read about our journey here.