





















I served as an area chair again for ICML 2026. In this blog post, I would like to share my experience this year.
This year, I got assigned 16 papers, which is substantially more than the number of papers I got assigned in any conference I served in any previous years. The workload is higher than I expected. I would have to squeeze more time from my daily work and weekends to do the volunteer work. To be honest, I found it hard.
This year, ICML 2026 implemented a two-policy framework for LLM use in reviewing:
Policy A (Conservative):
Policy B (Permissive):
Reviewers declare which policy they want to follow, and authors declare whether they require their papers to be reviewed under Policy A, or allow them to be reviewed under Policy B. Any reviewer who is an author on a paper that requires Policy A must also be willing to follow Policy A. Submissions are matched with compatible reviewers.
To detect LLM use in reviewing, ICML 2026 has modified submitted PDFs with a machine-readable watermark that will instruct LLM to produce specific phrases in reviews. If any review is flagged for LLM policy violation, it will be manually confirmed. The papers from the reviewers who violate the policy will be desk rejected, and there is one such paper among the papers assigned to me.
What’s somewhat confusing to the area chairs is that the area chairs do not know which policy each paper follows, but the reviewers themselves do know because reviewers declare their policy preference that should match the paper’s policy. For some reason it is intended by ICML 2026.
Among the 16 papers assigned to me, I recommended accepting one paper. One interesting fact is that I think I have been assigned with papers from the same laboratory or team that submitted this recommended paper in the past few years in a few different conferences and usually they are of good quality.
此内容由惯性聚合(RSS阅读器)自动聚合整理,仅供阅读参考。 原文来自 — 版权归原作者所有。