AISTATS 2018 Accepted Papers
Any typos will be corrected in the final list of proceedings. This is a temporary list in alphabetical order of the title. To address serious typos, please contact publicity chair Aaditya Ramdas at aramdas [at] berkeley.edu.
Contributions of AISTATS papers can be categorized into four areas a) algorithmic, b) theoretical, c) unifying or d) application.
Algorithmic contributions may make a particular approach feasible for the first time or may extend the applicability of an approach (for example allowing it to be applied to very large data sets).
A theoretical contribution should provide a new result about a model or algorithm. For example convergence proofs, consistency proofs or performance guarantees.
A unifying contribution may bring together several apparently different ideas and show how they are related, providing new insights and directions for future research.
Finally, an application contribution should typically have aspects that present particular statistical challenges which require solution in a novel way or through clever adaptation of existing techniques.
A paper may exhibit one or more of these contributions, where each of them are important in advancing the state of the art in the field. Of course, at AISTATS we are also particularly keen to see work which relates machine learning and statistics or highlights novel connections between the fields or even contrasts them.
One aspect of result presentation that is often neglected is a discussion of the failure cases of an algorithm, often due to concern that reviewers will penalize authors who provide this information. We emphasize that description of failure cases as well as successes should be encouraged and rewarded in submissions.
When reviewing, bear in mind that one of the most important aspects of a successful conference paper is that it should be thought provoking. Thought provoking papers sometimes generate strong reactions on initial reading, which may sometimes be negative. However, if the paper genuinely represents a paradigm shift it may take a little longer than a regular paper to come around to the author's way of thinking. Keep an eye out for such papers, although they may take longer to review, if they do represent an important advance the effort will be well worth it.
Finally, we would like to signal to newcomers to AISTATS (and to machine-learning conferences generally) that the review process is envisioned in exactly the same spirit as in a top quality journal like JRSS B, JASA, or Annals of Statistics. Accepted contributions are published in proceedings, and acceptance is competitive, so authors can rightly include these contributions in their publication list, on par with papers published in top quality journals. Further, AISTATS does not give the option to revise and resubmit: if a paper cannot be accepted with minor revisions (e.g., as proposed by the authors in their response to the reviews), it should be rejected.
Given the culture gap between the statistics and machine learning communities, we thus want to emphasize from the start the required levels of quality and innovation. All deadlines are very strict, as we cannot delay an overall tight schedule.
Confidentiality and Double Blind Process
AISTATS 2017 is a double blind reviewed conference. Whilst we expect authors to remove information that will obviously reveal their identity, we also trust reviewers not to take positive steps to try and uncover the authors' identity.
We are happy for authors to submit material that they have placed online as tech reports (such as in arXiv), or that they have submitted to existing workshops that do not produce published proceedings. This can clearly present a problem with regard to anonymization. Please do not seek out such reports on line in an effort to deanonymize.
The review process is double blind. Authors do not know reviewer identities, and this includes any authors on the senior program committee (i.e., the area chairs). However, area chairs do see reviewer identities. Also, during the discussion phase reviewer identities will be made available to other reviewers. In other words, whilst the authors will not know your identity, your co-reviewers will. This should help facilitate discussion of the papers.
If a reviewer requests code from the authors, this code should be anonymized (e.g., author names should be removed from the file headers). That said, we understand that it might be difficult to remove all traces of the authors from the files, and will exercise reasonable judgment if innocent mistakes are made.
The AISTATS reviewing process is confidential. By agreeing to review you agree not to use ideas, results, code, and data from submitted papers in your work. This includes research and grant proposals. This applies unless that work has appeared in other publicly available formats, for example technical reports or other published work. You also agree not to distribute submitted papers, ideas, code, or data to anyone else. If you request code and accompanying data, you agree that this is provided for your sole use, and only for the purposes of assessing the submission. All code and data must be discarded once the review is complete, and may not be used in further research or transferred to third parties.
The CMT Reviewing System
The first step in the review process is to enter conflicts of interests. These conflicts can be entered as domain names (e.g., cmu.edu) and also by marking specific authors with whom you have a conflict. The use of double blind reviewing means you may not able to determine the papers you have a conflict with, so it is important you go through this list carefully and mark any conflicts. You should mark a conflict with anyone who is or ever was your student or mentor, is a current or recent colleague, or is a close collaborator. If in doubt, it's probably better to mark a conflict, in order to avoid the appearance of impropriety. Your own username should be automatically marked as a conflict, but sometimes the same person may have more than one account, in which case you should definitely mark your other accounts as a conflict as well. If you do not mark a conflict with an author, it is assumed that you do not have a conflict by default.
CMT also requests subject information which will be used to assist allocation of reviewers to papers. Please enter relevant keywords to assist in paper allocation.
You can revise your review multiple times before the submission. Your formal invite to be a reviewer will come from the CMT system. The email address used in this invite is your login, you can change your password with a password reset from the login screen.
Supplementary material is allowed by AISTATS 2018. For example, this supplementary material could include proofs, video, source code or audio. As a reviewer you should feel free to make use of this supplementary material to help in your review, though reviewing supplementary material is up to your discretion. One exception is the letter of revision from papers previously submitted to NIPS. If this letter is present in the supplementary material, we ask you to take it into consideration.
Simultaneous submission to other conference venues in the areas of machine learning and statistics is not permitted.
Simultaneous submission to journal publications of significantly extended versions of the paper is permitted, as long as the publication date of the journal is not before May 2017.