blog.icml.cc web signal

ICML 2026 lays out 12 curated socials across three evenings

TL;DR

  • ICML 2026 will run 12 curated social events, four each on Monday, Wednesday, and Thursday, all from 7 PM to 9 PM.
  • Social Chairs Kevin Leyton-Brown and Chulhee Yun picked the lineup, spanning AI co-scientists, robot learning, cooperative AI, and decentralized training.
  • Access is frictionless: no waitlist or separate signup, but Monday socials require tutorial-day registration rather than the main conference pass.

The list of socials a conference picks is usually a better read on where a field's actual attention sits than the keynote lineup, and the ICML 2026 shortlist published on the ICML blog is worth a slow read for that reason. Twelve curated evening events, four each on Monday, Wednesday, and Thursday, all running from 7 PM to 9 PM.

The topics themselves are the part to sit with. Monday leans research-and-community: AI co-scientists in the research loop, an ICML chess club, an India @ ICML night, and AI for games. Wednesday tilts practical: how to network at an AI conference, robot learning, a bridge between research and the ML open-source community, and one on data foundations. Thursday is where the more pointed conversations live: AI negotiation and compensation, a Muslims in ML social, cooperative AI, and one titled 'Collective, Decentralized Training as a Hedge Against AI Power Concentration.' Those last two, sitting on the schedule as first-class socials, read as a small signal about what a chunk of the ICML crowd is now willing to put plainly in the program.

Social Chairs Kevin Leyton-Brown and Chulhee Yun did the curating, and access is deliberately frictionless. As the announcement puts it, 'if you're registered for ICML, you're welcome to that day's socials.' The one asterisk is that Monday's socials sit inside tutorial day, so you need tutorial registration to be on the floor for those; Wednesday and Thursday run under the main conference pass.

The honest caveat is that this is the announcement, not the write-up. It gives topics and rooms but not organizer names for the individual socials, no speaker lists, and no indication of how the 12 were picked out of what was almost certainly a longer pool. If you were hoping to know who is actually running the decentralized-training or cooperative-AI conversations, you will have to wait until closer to the conference.

What is useful in the meantime is the shape of the list. If you are a researcher, a recruiter, or someone tracking where the community's own attention is drifting, the socials give you a rough map of the corridor conversations that will not show up on the paper track.

Shared on Bluesky by 2 AI experts