← sources

ICLR27 math and stats paper review experience [D]

source: reddit (reddit:MachineLearning)published 2026-11-07 03:02ingested 2026-11-08 02:37

retained text

For people who had papers submitted to ICLR that are applied theory style. For example, 60- 70% theory with some experiments at the end. What was your review experience like for ICLR27?

We got a score of 222, on the probabilistic methods track.

All 3 reviewers attacked our novelty and asked for additional experiments. Not a single word was mentioned about our theory, which is our biggest contribution. One outright said he didn't understand this paper or why this task exists. It's a well-defined task with lots of prior work.

I used accept-odds, tracked submissions that were similar to mine. It seems like they have all gotten similar comments. Not a single comment on the math, all comments were on the experiment. A paper with 20 pages of proofs even got comments:

>Assuming the proofs are correct (which can be done by AI in 2026), the theorems only show trivial conclusions.  
Rating: 1: Clear rejection

I am starting to suspect, because there were no bidding this year. Many theory-heavy papers got sent to pure comp sci reviewers.

Would be nice if people can share their experience, I am trying to work out if I should engage in the rebuttal or just withdraw.

document fetched 2026-11-08 00:24 · canonical https://reddit.com/r/MachineLearning/comments/1xpvtxt/iclr27_math_and_stats_paper_review_experience_d