NeurIPS: A Beginner’s Guide

NeurIPS 2018 has recently concluded, and I was lucky enough to have been able to attend. As a student volunteer and first time attendee, I didn’t have a paper to present, but also didn’t have to deal with the 11 minute 38 second window in which you could buy tickets. Instead I worked one day checking people in, and for the rest of the conference just roamed around feeling very noobish. I had lots of questions and anxieties about what was gonna happen going into the conference, so I’ve written some things that I would’ve liked to have heard beforehand. In the follow blog post, I’ll define the different events that happen at the conference, as I didn’t know what they were before I arrived, and then give you some Do’s and Don’ts that defined my experience. It ended up being a little long, and while I did try to make all the content relevant, there’s a TL;DR for those with shorter attention spans or who just ain’t got time for all that (no offense taken).

Events:

Workshops: Showcases that are curated by people other than NeurIPS. They can be organized by either a company, a marginalized group, like “Black in AI” or “WiML,” or by topic, for example “Continual Learning” or “Creativity in ML.” These all have individual schedules that you should check on the day they occur. Primarily grouped together at the end of the conference, they mostly co-occur, so you won’t be able to go to all of them.

Tutorials: Someone talks at you about a topic for one or two hours. All tutorials were on the same day, but I missed them as I had to volunteer.

Poster Session: Two rooms are filled completely with poster versions of submitted papers. The authors, if it’s a large paper not usually all of them, stand by the posters to give overviews, field questions, and respond to thoughts about their work.

Parallel Tracks: Quick (5 to 15 minutes) oral presentations of select papers that are in the following poster session. Usually three of these tracks are going at the same time and are loosely grouped into categories.

Demos: specific demonstrations of applications. I saw chatbot avatars showcased, large inference chips demonstrated, and data labeling tools explained. These are usually companies showing off a product they’ve made.

Quick Note on Volunteering: While the listing asks you to be a current student, I graduated in May 2018 and was able to attend the NeurIPS conference that occurred in December 2018. As I had been a full-time student in the same calendar year, I was allowed to apply.

Do’s and Don’ts:

Do a little preparation. Not a huge amount if you’re just attending, but at least familiarize yourself with a few papers that seem cool and then go to the poster for clarifications. If you’ve been tagging along with the constant flow of ML papers over the year, you’ll stumble across a few papers you’ve already read. I suggest mapping out a few important posters you want to see as well as some workshops that interest you, but a list of questions to ask researchers will go way further than a detailed schedule (which would change once you get there anyway).

Do go around to the company booths. Many give out free swag while begging you to work for them. Definitely try to talk to both a recruiter and an engineer for each company. It felt to me the recruiters were a little clueless about the actual work that’s done there, while the engineers were a little clueless on how to get hired at that company. If you’re looking for work, this will be where you pick up the business cards that skip your resume past the online portals. And often the recruiters at the booths will let you know about certain exclusive networking events…

Do seek out those networking events, and sign up for things when you get emails about them before the conference. Doesn’t matter what it is - just sign up for it. Mentoring breakfasts, resume reviews, dinner, some reception, whatever. They’ll help. (At the very least it’ll be free food.) The parties that occur at nearby bars or museum lobbies after hours are massively beneficial for meeting people, networking, and applying to jobs. It’s also the best opportunities to ask more candid questions. I recently graduated from undergrad, so moving into the tech world is gonna dramatically shift my social life as well. Clearly it makes more sense to ask about that over drinks than stopping someone during a poster session. At one party, I traded a drink ticket for a referral to the company. I don’t even drink. Muahahahaha. I’ve repeatedly heard, and felt this myself, that the benefit of the conference is not understanding the 1000 posters (because you won’t be able to), but rather the connections (networking) you make in person.

Do talk to your idols. They’re extremely human, and also super willing to talk to you. At no point in the conference did it occur to me that I “wasn’t important” enough or “too new” to talk to anyone. Jürgen Schmidhuber, Jeff Dean, Ian Goodfellow, and a slew of other influential AI people were all super kind and engaging. Sometimes all you need is “Hey, I follow you on twitter!” My personal favorite was just “Hell yeah, LSTMs!” to Jürgen to start a conversation. Ask them for advice, their opinions on super grand AI problems, anything. The hardest part will be getting a chance to butt into the conversation, which leads to my similar yet slightly different next point….

Do think about the conversations you have. You’ll get lots of practice with non-networking social things like saying people’s names, starting conversations, backing out of conversations, inserting yourself into a conversation, splitting off from someone when you’re interested in different topics, and lots of other stuff that’s hard to pinpoint exactly. You’ll spend the week talking to people you don’t know but with whom you share many interests, which is perfect for conversation. You really don’t need much more than “what are you here at NeurIPS for?” to start a talking. Also, when you ask people questions, make sure you listen to their reply and sum it up to yourself in your head afterwards. It’ll help it stick a lot more.

Do make friends. I’m only 22, so meeting a few people my age helped make my week in Montreal more enjoyable while meeting some older people from the field gave me some wisdom. Always easier said than done but having someone to get meals with will make things easier. I stayed in a hostel, and it was pretty easy to see who was there for the conference and who wasn’t. Just ask about the conference, and viola! Now you have someone to go to and from the venue with and maybe even talk about the day after you’re back in your room.

Do explore the city. Montreal was beautiful, and some of my favorite parts of the conference occurred outside the conference center. You’ve traveled this far, I highly suggest doing a little sightseeing. Vancouver, the NeurIPS location for the next two years, will be plenty beautiful as well.

Do download the app and use it. It wasn’t just convenient, but actually generated me opportunities. I used the Whova app to meet up with people for lunch several times, schedule small meetups during coffee breaks with people whose work I was interested in, and keep track of the schedule. Feel free to just send people messages, it’s not obtrusive, if they don’t want to talk they just won’t respond.

Do ask ethical questions. Don’t harass people or be rude about it. There’s a myriad of ethical problems giant tech companies have. Algorithmic bias, data privacy, algorithmic abuse (fake news) are all relevant to Machine Learning researchers. Debate the researchers, don’t harass an intern.

Don’t stay at the conference every second it’s going on. You will get burnt out. A lot of the “don’ts” in my list boil down to advice on not burning yourself out over the course of the week. By the end of the final workshop Sunday, I had to sit down and just be on my own for an hour before the reception. NeurIPS is a huge dose of intellectual thinking, socializing (not much alone time), and just generally being occupied the whole week. It’s okay to skip a talk or a set of oral presentations that doesn’t concern you and relax in your hotel room or roam the city. In the same vein….

Don’t get FOMO. You’re never gonna be able to absorb everything, and definitely don’t get down about that. There are posters, events, and parties going on the whole week, you’re bound to miss stuff. Don’t feel bad if you spend a night in to recoup, if you eat a lunch alone, or if you really really wanted to go to the NVIDIA party but didn’t get a wristband. If you miss the event for that specific company you want to work for, you can catch up by standing around the booth the next day and talking to all the researchers there. Stuff will happen around you that you will miss, but there’s also so much happening that missing something specific will not mean you miss “the one connection” that lands you the job. Also, many of the talks are posted online afterwards, and almost always at least the slides.

Don’t scrimp on your sleep. Being cognizant for the entire day is worth missing the first talk. The company parties/networking events can keep you out until midnight, and the first talks are at 8:30 everyday. With that being said, the invited speakers were some of my favorite parts of the conference. Picking and choosing what’s most important is what makes the experience yours, but don’t always choose events over sleep. Went back to my hostel and took a fat nap one day and that was super legit.

Don’t get overwhelmed. NeurIPS is “end game content,” in that people put a lot of effort into this work, and that works starts with their PhD’s or MSc. When you spend all day looking at the most cutting edge research in Artificial Intelligence, it can be easy to think, “Damn I’m never gonna get to this level.” What they present is the accumulation of months of hard work. If you’re a first time attendee as a student, most of it is gonna be beyond your scope (like you’re not gonna know what most of the words in the poster titles are), which is likely to be true even if you’re presenting. Imposter syndrome can hit hard, but you’ll get there eventually. Another repeated piece of advice I’ve heard about machine learning is that it’s not the most clever people that succeed in this field, it’s the most persistent.

TL;DR: The first two sentences of each Do and Don’t act as summaries for that point. With that being said: Don’t get burnt out, it’s a long conference. Go to the networking events and ask around for them at the booths. Make some friends, talk to people you think are impressive, talk to people who might be impressive but you don’t know yet, as they will also be impressive. Bring a list of questions and study a couple papers beforehand. Sometimes it feels like everyone else is way better than you at neural networks - don’t worry, they’ve practiced lots. You can get there too.