ECCV 2026 lists E.T. workshop on representation learning theory

Buried in the ECCV 2026 program is a workshop called E.T., short for Empirical Theory in representation Learning, listed on the conference schedule with sessions running from 8 a.m. to noon and 2 p.m. to 6 p.m. on Tuesday, September 8 in Malmö, Sweden. The workshop's OpenReview page is live but currently reads as a placeholder, without a public program or submission details visible.

The framing is what makes it worth flagging early. Representation learning sits under a lot of modern computer vision, from self-supervised pretraining to the visual encoders inside multimodal systems, and pairing 'empirical' with 'theory' in the name signals a venue that wants theoretical claims tested against actual vision experiments, rather than pure math on one side or pure benchmark chasing on the other. That is a niche the main ECCV track does not always accommodate.

The honest caveat is that the listing does not give you the substance yet. There is no organizer list, no invited-speaker roster, no call for papers, no deadline, and no word on whether accepted work will be archival. The OpenReview stub is the sort of placeholder that typically goes up months before a call actually opens, so anyone considering a submission is waiting on the same information you are.

For researchers doing early-stage theoretical work on contrastive objectives, scaling behavior, or vision encoders, this is one to watch when the full call is posted. The value of a workshop like this is usually the discussion in the room, and the ECCV crowd is exactly the practitioner mix a theory-of-representations venue most needs to pull in.