The International Conference on Machine Learning (ICML) is the oldest of the major machine learning conferences and, together with NeurIPS and ICLR, one of the three highest-impact venues in the field. It began as the International Workshop on Machine Learning at Carnegie Mellon University in Pittsburgh in July 1980 and has run as a full annual conference since 1993, organized by the International Machine Learning Society (IMLS). Unlike a journal publisher, ICML has no journals, no subscriptions, and no article processing charges: accepted papers are published open access in the Proceedings of Machine Learning Research (PMLR), free for anyone to read, with authors retaining copyright.

This guide covers how publishing at ICML actually works: whether your paper fits, the double-anonymized OpenReview process, the annual timeline, formatting, costs, and how ICML papers are cited and indexed.

ICML runs on a fixed annual cycle rather than rolling submission. For ICML 2026 (July 6-11, Seoul): abstracts were due January 23, 2026, full papers January 28, reviews landed in March, and authors were notified April 30. Accepted papers are presented at the conference and published as a volume of PMLR - ICML 2025 (Vancouver) is PMLR Volume 267.

Key structural points:

• One shot per year. Miss the January deadline and the next ICML is a year away - most ML researchers cycle between ICML, NeurIPS, and ICLR deadlines.
• Conference paper format. Eight pages of main content (ICML 2026 rules), with unlimited pages for references, impact statement, and appendices; camera-ready versions get one extra page.
• Dual-submission rules. You may not submit work that is identical or substantially similar to papers published, accepted, or under parallel review elsewhere. Preprints on arXiv are permitted - the anonymity policy governs how you handle them.
• Reciprocal reviewing. Authors are expected to serve as reviewers; ICML 2026 requires it for submitting authors.

• Model: double-anonymized (double-blind). Submissions must be fully anonymized; violations are grounds for desk rejection.
• Platform: OpenReview (ICML moved from CMT to OpenReview in the early 2020s). Unlike ICLR, ICML reviews are not fully public during the process.
• Process: reviewers score the paper, authors respond in a rebuttal phase, area chairs make recommendations, and program chairs finalize decisions.
• Outcomes: acceptance as poster, spotlight/oral (top papers), or rejection. Acceptance rates have ranged roughly 21-30%; ICML 2026 reportedly received about 24,000 submissions with ~26.6% accepted.
• Timeline: about three months from the paper deadline to notification.

1. Read the Call for Papers at icml.cc for the current year - scope, page limits, and policies change year to year (including current rules on LLM use in writing and reviewing).
2. Prepare your manuscript with the official ICML LaTeX style file (e.g., icml2026.zip). SciSpace also hosts the ICML format for one-click formatting.
3. Anonymize - remove author names, affiliations, acknowledgments, and identifying links.
4. Register the abstract by the abstract deadline (about a week before the paper deadline), then submit the full paper via OpenReview.
5. Engage in the rebuttal phase - author responses can and do change scores.
6. If accepted: de-anonymize, apply camera-ready formatting with the extra page, sign the PMLR publication agreement, and register at least one author to present in person or as required that year.

Publishing at ICML is free. There is no submission fee and no APC. The real costs are conference registration and travel for the presenting author - typically several hundred dollars for registration, with reduced student rates and financial assistance programs. Every accepted paper is published open access on proceedings.mlr.press at no charge, and authors retain copyright. Because PMLR is fully open, there is nothing for transformative agreements or institutional deals to cover.

ICML papers use author-year (natbib-style) citations in the ICML template. Cite published papers as: Author(s). Title. In Proceedings of the 42nd International Conference on Machine Learning, PMLR 267, 2025. PMLR volumes carry ISSN 2640-3498, and each paper has a stable PMLR landing page with BibTeX. ICML proceedings are indexed in the usual CS discovery layers (Google Scholar, DBLP, Semantic Scholar) and PMLR is interlinked with JMLR.

ICML operates a code of conduct for participants and a publication-ethics framework covering plagiarism, duplicate submission, and (recently) explicit rules on LLM-generated text and prompt injection in submissions. Papers must include an impact statement discussing broader consequences. An ethics-review track can escalate flagged submissions to dedicated ethics reviewers. Camera-ready papers that violate policies can be removed from the proceedings.