The International Conference on Learning Representations (ICLR) is the premier conference dedicated to deep learning and representation learning, and one of the three highest-impact venues in machine learning alongside NeurIPS and ICML. It was established in 2012 by Yoshua Bengio and Yann LeCun, with the first conference held in 2013, and from the start it has used an open peer-review model: submissions, reviews, author responses, and final decisions are all public on OpenReview.

Publishing at ICLR is unlike publishing with a journal publisher. There is no journal portfolio, no APC, and no subscription paywall - a single annual proceedings, entirely free to read, with acceptance decided through a fixed yearly review cycle. This guide covers what ICLR publishes, how the OpenReview process works, formatting and citation requirements, costs, and the ethics policies you agree to when you submit.

ICLR publishes one annual conference proceedings hosted on OpenReview, covering representation learning in the broadest sense: feature and metric learning, generative models, optimization for deep learning, reinforcement learning, and applications across vision, natural language processing, speech and audio, robotics, and neuroscience. Recent editions have grown rapidly - ICLR 2025 accepted 3,704 of 11,603 submissions (about 32%), up from 2,260 of 7,304 (about 31%) in 2024.

Accepted papers fall into presentation tiers (oral, spotlight, poster), but all appear identically in the proceedings. ICLR also runs a workshop track each year for early-stage and focused work, and recent editions have added a Journal-to-Conference track that lets accepted papers from selected journals (such as TMLR) be presented at the conference.

Everything runs on OpenReview (openreview.net). The cycle is annual and the deadlines are hard - there is no rolling submission. For ICLR 2027, the abstract deadline is September 18, 2026 and the full-paper deadline is September 25, 2026, with reviews released in early November and final decisions in mid-December 2026.

1. Register on OpenReview. Institutional email addresses are strongly recommended; new-profile creation closes shortly before the deadline.
2. Submit an abstract by the abstract deadline, then the full paper about a week later.
3. Reviews are posted publicly (reviewers stay anonymous), typically 6-7 weeks after the paper deadline.
4. Author discussion period. You respond to reviews and can revise the PDF during the rebuttal window.
5. Decisions are public. Accept/reject decisions and meta-reviews appear on OpenReview; accepted papers are immediately part of the open proceedings.

Submissions are double-anonymized: reviewers cannot see author names and authors cannot see reviewer names. Posting a preprint on arXiv is explicitly allowed under the dual-submission policy, but the submitted PDF itself must be anonymized and self-citations written in the third person.

ICLR pioneered open peer review at scale, based on a model proposed by Yann LeCun. Every submission - including rejected ones, unless withdrawn early - remains publicly visible on OpenReview together with its reviews, scores, author responses, and the area chair's meta-review. Anyone can read the discussion, and the community can post public comments. This transparency is ICLR's defining feature: you can inspect years of real reviews before submitting, which is the best calibration tool available for judging whether your paper fits.

• Template: official ICLR LaTeX style files (e.g., iclr2026.zip from the ICLR Master-Template repository); a matching template is available on SciSpace.
• Page limit: 9 pages of main text at submission for ICLR 2026, expandable to 10 pages for the camera-ready; references and appendices do not count toward the limit.
• Citations: author-year natbib citations - the style file sets authoryear with round brackets, so use \citep and \citet rather than numbered references.
• Supplementary material: code and appendices may be uploaded by the same deadline; reviewers are not obliged to read them, but code submission is encouraged for reproducibility.

There is no fee to submit or publish at ICLR. Accepted papers are published open access on OpenReview at no cost; authors retain copyright, and papers are commonly released under a CC BY license (check the current year's camera-ready instructions for the exact terms). The only costs are attendance-related: at least one author must register for the conference to present, and registration fees plus travel are the real budget line. ICLR 2026 was held at the Riocentro Convention Center in Rio de Janeiro, Brazil, on April 23-27, 2026; ICLR 2027 is scheduled for April 26-30, 2027 in California.

ICLR maintains a code of ethics and a code of conduct that all authors and reviewers agree to. Recent editions added explicit policies on the use of large language models in both papers and reviews - undisclosed LLM ghost-writing of reviews or papers is sanctionable. Plagiarism, dual submission to another archival venue, and identity-revealing content in anonymized submissions are grounds for desk rejection. Reviewer misconduct and collusion-ring behavior are actively investigated, aided by the fully public review record.