Inundated AI conferences impose submissions caps

In a bid to tackle a surge in submissions, conferences and journals focused on artificial intelligence are introducing caps on the number of papers that a researcher can submit.

The 2027 International Conference on Learning Representations (ICLR), which will take place next year in San Francisco, California, has announced that authors will be capped at 20 submissions each. Authors who have never had a paper accepted at a major AI conference will be limited to one submission, unless they have a co-author who meets the authorship requirement.

The move comes as the number of submissions to ICLR 2027 looks set to exceed 62,000, up from 490 a decade ago. “Historically, the vast majority of new researchers have focused their effort on a single paper or project, and we want to incentivize this practice, rather than encouraging people to submit very large numbers of, usually AI-generated, papers,” says Jacob Andreas, a computer scientist at the Massachusetts Institute of Technology in Cambridge and ICLR 2027 programme chair.

The 2027 Annual Conference on Artificial Intelligence and Statistics, to be held in Montreal, Canada, in May, has introduced a cap of 12 submissions per author, with those who haven’t had an accepted paper at a major AI conference capped at 3 submissions. The Association for Computational Linguistics has set a cap of 20 submissions per person for its conference in Budapest later this month and 5 for first-author papers per review cycle, citing the “unsustainable rate of submissions”.

Not everyone agrees with the policies. In response to a LinkedIn post about the ICLR changes, some researchers pushed back on the perceived double standard. “If a productive PhD student has two papers to send, why stop them and allow 20 submissions from a research lab with prior ‘major conference’ papers?” said Aylin Yener, a computer scientist at Ohio State University in Columbus, Ohio.

Abhinav Aggarwal, a research engineer at technology firm Meta in Seattle, Washington, had similar criticisms, calling the restrictions on early-career researchers “very disappointing”.

Kathleen Fraser, a computer scientist at the University of Ottawa in Ontario, Canada, said that she can “see the logic” in the new rules, which limit the number of papers from people who cannot contribute to the review process. The ICLR rules state that only authors who have had a paper accepted at a major AI conference can review submissions.

Testing approaches

Gautam Kamath, a computer scientist at New York University who started the LinkedIn discussion, thinks that the cap of 20 submissions might be too high, but notes that it’s important to cater for large research groups that list the laboratory leader on every paper.

Kamath is co-editor-in-chief of the AI journal Transactions on Machine Learning Research, which in June implemented new quotas on submissions, after a tripling of submissions over the past year. This included a 13-fold increase in single-author submissions.