ICLR 2018 Kicks Off in Vancouver

AI researchers are gathering in Vancouver, Canada for the sixth annual ICLR (International Conference on Learning Representations). The event starts today and runs to May 3rd at the Vancouver Convention Centre.

Compared to NIPS, ICML, and ACL, the ICLR is a relative newcomer. Founded in 2013 by deep learning mavericks Yoshua Bengio and Yann LeCun, the conference has become an important destination for AI researchers.

Program chair Yann LeCun tweeted that “attendance at ICLR doubles every year. It’s up to 2000 this year. It merely doubles because attendance is capped. The number of submissions also doubled from last year. [Google Researcher] Tara Sainath is opening the conference.”

This year 337 out of 935 paper submissions were accepted, with best paper honours going to On the convergence of Adam and Beyond, Spherical CNNs and Continuous Adaptation via Meta-learning in Nonstationary and Competitive Environments.

A statistical analysis by Arthur Pajot shows Google leading both the paper submission and acceptance charts, ahead of academic institutions Berkeley, Stanford, CMU, UofT, and ETH; and corporate research teams from Facebook, Microsoft and IBM. Google is dispatching 130 researchers to the conference and pitching in US$40,000 as a platinum sponsor.

The most prolific individual submittors are Yoshua Bengio, who submitted 18 papers, (9 rejected, 7 posters, 2 workshops), and UC Berkeley’s Pieter Abbeel with 12 submissions (4 rejected, 4 posters, 2 workshops, 2 orals). The top-submitting countries are the United States, United Kingdom, China, Canada, and Germany.

A paper keyword analysis conducted by Pau Rodríguez suggests “meta-learning, exploration, model compression, adversarial examples, variational inference are the hottest topics this year. For instance, 85% of the papers with the keyword exploration were accepted, while classification and CNN only show 12% acceptance rate.”

The ICLR is known for popularizing the open review system designed by Andrew McCallum from UMass Amherst. Papers submitted to the conference are published on http://OpenReview.net, publicly reviewed and archived on the site.