Discussion Articles in JMLR

Beginning in September 2006, the editorial board of the Journal of Machine Learning Research will invite commentary and discussion of select papers, and will publish this discussion in the Journal.

The published discussion will include short invited comments that discuss the content and contributions of the paper, highlighting its place in the literature and potential impact on the field, while raising interesting and provoking questions. The discussion will be restricted to a small number of papers each year, and is intended for papers that contain highly novel, important, or surprising results which are expected to have a significant impact on the field. The discussants will be invited by the JMLR action editor in charge of the paper, and the commentary will be edited and approved by the editors before publication. The discussants will typically be well known, often senior researchers in the field, chosen for their familiarity with the body of work related to the paper, and for their potential to provide a variety of opinions and perspectives. Between five and ten discussants will be invited for each paper, and the authors will be invited to preview (but not veto) these comments and write a brief rejoinder.

Such discussion has successfully appeared for many years in journals such as the Annals of Statistics and Statistical Science. Some of the benefits are: (1) discussion can bring in different points of view and can help place a paper in a broader context (2) the discussion can be instructive and educational, in particular for students (3) discussion can broaden the machine learning community by having discussants who might not typically publish in JMLR, leading to new involvement and perspectives (4) the discussion can sometimes become of even more interest than the paper, and is given separate citation (5) the discussion can sometimes be quite entertaining, especially when it gets heated.

The goal is to provide a venue for opinions, reactions, judgments of importance, and links/connections to other ideas and subfields, all of which have largely been absent from the Journal until now. By imposing a fairly tight schedule on the discussants, and staging the publication of the paper and the discussion, a discussion article will be reviewed and published without an increase in the time to publication. The editor will appoint one or more discussion editors to service the extra load that discussion articles will entail.

The logistics of the review cycle for a discussion article are as follows:
  1. A submitted article enters the review cycle in the usual manner: an action editor is assigned, and (typically) three referees are recruited.
  2. At the point where the paper is accepted (which may in some cases be before all of the reviews are in), the action editor may nominate the paper to the editor as a discussion article.
  3. The editor solicits comments on the paper from the editorial board. The editor and discussion editors review these comments and make a decision in favor or against making it a discussion paper.
  4. If the editor decides to select the paper as a discussion paper, the authors are contacted; they have the option to decline to have the paper discussed.
  5. After the paper is selected as a discussion paper, the action editor recruits between five and ten discussants, giving them access to the paper via the JMLR system. The discussants may include some of the referees of the paper.
  6. Each discussant writes a commentary, typically between one and three pages, but occasionally longer. There is a four week deadline to submit the commentary to JMLR, although the action editor may extend or shorten the deadline. While commentary is being gathered, the usual procedure of revisions and review moves forward with the article, so that its publication is not held up by the discussion component.
  7. After receiving each discussion piece, the action editor may ask for minor revisions, for example asking the discussion to be shortened. The editors (including the action editor, discussion editors, and editor-in-chief) reserve the right to reject any commentary. However, this will only occur when the commentary is irrelevant or overly self-serving, abusive, or otherwise inappropriate for publication in a scientific journal, in the judgment of the editors. Once all the commentary is received, or at a deadline imposed by the action editor, all commentary is sent to the authors. The names of the discussants are known to the authors at this point.
  8. The authors are asked to return a brief rejoinder to the editors after a short time, typically two weeks. The authors do not have veto power over the comments; if they disagree with a comment or feel that a comment misrepresents some aspect of their paper, these concerns may be addressed in the rejoinder.
  9. In parallel with the discussion being organized, the final version of the paper is prepared and published by JMLR. The final discussion is published separately, as close to the original article in issue/page number as possible. The discussion will be citeable, by title of the article and each discussant's name.

Copyright © JMLR 2000. All rights reserved.