Double submission guide

Many people are conflicted about publication venues, wanting to publish in multiple ways. This is your guide to conflict resolution for submissions to ICML. Find the exact case you are considering below, and follow the directions. (There are subtle but important differences between every case below.) If you do not find your case below, email us.

For the following, two papers will be considered the same if 50% or more of the novel aspects of an ICML submission are in separate papers.

To Journal

You submit a paper to ICML on February 24. On February 25 (or after), you want to submit a long form draft of the paper to a journal.

This is ok, so long as you follow all constraints for the submitting journal. This paper is reviewed as a normal ICML paper and you need not inform ICML.

ML Journaled

You published the paper at a machine learning journal (MLJ or JMLR are allowed), and have not presented the material at a machine learning conference.

You may submit the paper to ICML “Not For Proceedings”. You should prepare a short abstract with a pointer to the full paper. Your paper will be considered for presentation at ICML, but not for inclusion in the official proceedings. This consideration is not a formal review —acceptance is based on a subjective judgement of audience interest. In essence, we will assume the correctness of all material and judge interest as a normal ICML paper.

Journaling or voluntary NFP

Your paper is under submission (before Feb 24) at a journal, and not yet published. Or, you simply do not want your paper to appear in the official proceedings of ICML.

You may submit an ICML format paper to ICML “Not for Proceedings”. You must fit the paper into ICML format, to make it suitable for a quick review. Your paper will be considered for presentation at ICML, but not for inclusion in the official proceedings. This paper is reviewed as a normal ICML paper.

Published elsewhere

You have a paper on machine learning technology of plausible interest to people at ICML published at a venue with little audience overlap. We specifically exclude AIStats, COLT, KDD, NIPS, and UAI. This category is also not for for workshops.

These submissions will be held to a higher standard in general, as they must meet a test of interestingness for conference participants even though they have been or will be published elsewhere. This test is impossible to satisfy for conferences with strongly overlapping participants. Any submission judged a “weak accept” under standard ICML criteria will not be presented.

Legal double submissions

Publication in venues that do not review for novelty and correctness or do not have an official proceedings is allowed.

Publication at http://arxiv.org explicitly does not conflict with ICML. Presentation at a workshop which does not have both a review for novelty and correctness and an official proceedings is explicitly allowed.

Virtually all workshop submissions are ICML-compatible. If a workshop is a long running series with reviewed submission, presentations, and an archival status, then it is considered a conference under these criteria.

Illegal double submissions

You want to submit to another conference and ICML.

This is not allowed. You may not simultaneously have a paper under review at ICML and under review for a different public presentation venue.

Two examples of workshops that count as conferences are: (1) The Workshop on the Algorithmic Foundation of Robotics (workshop proceedings published in a book by Springer). (2) The European Workshop on Reinforcement Learning (selected papers published in LNCS).