Conferences: What is the Deal?

Sunday July 3, 2016

I was fortunate to attend all six days of the International Conference on Machine Learning in New York this year. I hadn't attended such a conference before. Here's what I learned about how academic conferences like ICML go down.

Conferences: The Deal

Tutorials: I usually think of tutorials as being mostly how-to affairs, possibly with a hands-on component. But ICML tutorials are just long talks. They run for two hours each, and provide fairly general overviews of a given sub-field or problem type. They're good for getting a feel for a broad area, with less focus on new work. The first day of the conference (Sunday) was given over to these.

Receptions/Parties/etc.: There's a ton of food and drink and people! There are official conference events, and then there are also company-sponsored affairs. There's a "job opportunities" email list you can get on when you register for the conference, but the real reason to subscribe is to get the invites for all the social events. It's exhausting.

Main conference: People give fifteen-minute talks about their new papers in giant rooms with barely any electrical outlets. This goes on for three days. Monday. Tuesday. Wednesday.

Poster sessions: These are what you really want to go to during the main conference days. You can quickly check out way more papers than you could sit through the talks for, and you can usually start talking to one of the authors immediately if you're interested. I found more cool things this way. I suppose if you were really proactive you could check out all the papers in advance online, but then why are you at the conference anyway? (That's a good question in general, by the way; see below.)

Workshops: The workshops are like whole mini-conferences, each a day long. These ran on Thursday and Friday. They succeed or fail with their organizers. I went to a great one on Thursday, and I visited four different ones on Friday. Many have their own poster sessions, which are great if you can hit them.

Conferences: A Good Deal?

Maybe you should stay home! With papers on arXiv and a lot of video turning up online, it may not be worth the trip and days of time, if all you want to do is learn about work in the field. Of course if you want to meet collaborators or look for a job, your incentives are quite different.

I'm sure other conferences are different. Even the conference that might be a "competitor" with ICML, NIPS, has quite a different feel, I'm led to understand.