ICLR

42,000 Submissions Push ICLR to Reconsider Its Acceptance Rate

42,000 submissions is a staggering number for any conference, and it forces ICLR's program chairs into a tough corner.

3 min readMachine Learning

The 42,000 submissions flooding ICLR this year force a decision that will define how seriously the field treats its own quality signal. Program chairs either hold the line on a ~26-28% acceptance rate, which would mean accepting over 11,000 papers, or they impose a hard cap and let the rate collapse. There is no third path that preserves both inclusivity and rigor.

The scale itself is the story. When a single conference receives more submissions than most journals process in a decade, the traditional acceptance-rate model breaks. Sticking to the historical percentage would produce a program so large that no attendee could meaningfully engage with it, and no reviewer panel could maintain consistent standards. The alternative, capping accepted papers at, say, 3,000, would yield an acceptance rate below 8%, transforming ICLR from a broad tent into an exclusive club. Both outcomes change what it means to have your work accepted there. This tension echoes themes we have explored in University Research Rankings Reveal AI Spreadsheet Innovation Trends, where the pressure to publish volume increasingly conflicts with the ability to verify quality. The same dynamic now confronts conference organizers directly.

Our view is that the chairs should cap the total number of accepted papers and accept the lower rate. Here is why: an acceptance rate is only meaningful if the denominator, submissions, is composed of genuinely novel work worth evaluating. At 42,000 submissions, that assumption is strained. Many papers are likely incremental, derivative, or hurriedly submitted to meet deadlines. Accepting 11,000 of them does not validate the field's growth; it dilutes the conference's signal. A hard cap forces the community to confront the real bottleneck, which is not how many papers deserve attention but how many can receive proper scrutiny. This is a practical matter of reviewer capacity, not an ideological stance on openness. As we noted in Clarity in your ICLR rebuttal can help reviewers see your revisions, the quality of the review process depends on reviewers having the time and focus to engage deeply. A program of 11,000 papers makes deep engagement impossible.

The practical consequence for researchers is straightforward: if you are submitting to ICLR this year, your paper faces a much higher bar than the headline acceptance rate might suggest. Do not assume the historical percentage will hold. Prepare your work as if the threshold has already moved, because even if the chairs keep the percentage, reviewer fatigue from handling so many papers will effectively raise the standard anyway. The specific detail to watch is the program chairs' announcement about reviewer allocation. If they reveal how many reviewers they have recruited and how many papers each will handle, that number will tell you more about the real acceptance bar than any rate ever could.

From Machine Learning

Given the massive jump to 42,000 submissions this year, I’m curious how people think the program chairs will handle the acceptance rate. Will they stick to the usual ~26-28% (meaning ~11k+ accepted papers), or will they cap the total number of accepted papers due to venue/reviewer constraints and drop the acceptance rate significantly?

Read the original at Machine Learning