12:30 - 1:45pm Tuesday, December 08, 2009
Stanley Room (34th floor)
Yann LeCun, New-York University
Fernando Pereira, Google Research
Daphne Koller, Stanford University
Lawrence Saul, UC San Diego
Tom Dietterich, Oregon State University
Leon Bottou, NEC Laboratories
Zoubin Ghahramani, Cambridge University
Terry Sejnowski, Salk Institute
Wolfgang Maass, Technische Universität Graz
Our community is vibrant but many questions, complaints, and suggestions for change have been voiced about the way in which the science we produce is evaluated and disseminated. This lunchtime debate and discussion is an opportunity for us all to explore possible changes to the current publication model in our community. The panel members are senior pillars of NIPS with editorial experience. Below are some of the questions that they would like to discuss with you. We would like this discussion to continue beyond the 75 minutes of the Tuesday debate, maybe electronically.
Issues with the current model:
- The conference/journal model in many areas of computer science like machine learning is very different from what happens in other communities: why is it that way in CS? as discussed by Lance Fortnow, is that the model we want? are there alternatives?
- Is author-anonymity a barrier to faster availability of papers? (because for conferences with anonymous submission the paper could otherwise have been on a public archive from the time the work is done to when the final paper is published). What about the 'rich get richer' problem if we do not have author anonymity?
- Reviewers are often feeling overloaded and not always very motivated.
- Should we question the current practice of confidentiality and anonymity of the reviews? i.e. publishing the reviews might yield better reviews, even if anonymous.
- There is a 'reviewer crunch' in July when most of the community has to review so many papers quickly.
- The differences between the CS publication model and the neuroscience publication model is a problem for the multi-disciplinary nature of NIPS, yielding a loss of the computational neuroscience contributions over the years.
Proposals for changes:
- Do the new electronic tools, archives, collaborative filtering etc. provide the opportunity for new and better ways of doing things?
- With modern technologies publications delays could be zero, but evaluation delays remain substantial. How do we leverage the above to make scientific communication more efficient?
- How do we reward the reviewers to motivate them better, e.g. by allowing the reviews to be published and cited?
- Can we change the submission and evaluation process to improve the quality of the selection of papers and the reviewer crunch? Maybe journals and conferences can team-up in order to find better solutions?
- How do we reconcile and ideally disentangle the conflicting roles of efficient retrieval, score-keeping, and stimulation that conferences and journals play?
Your opinion matters, please come and express it, participate in the debate. Do not hesitate to bring your sandwich!