CS conferences

The debate has flared up again, and I have only a little to contribute, but here it is.

Features of the present system, some of them very different from most other disciplines:

  • intense competition for acceptance into prestigious conferences
  • a very large number of conferences overall, most rather small
  • conference papers tend to be shorter with fewer details than journal papers
  • conference papers count as much, if not more than, journal papers
  • conference papers are not refereed as deeply as journal papers, but have to be acceptable to a broader audience
  • conference proceedings to some extent set the research agenda for the community

Negatives associated with these features:

  • too much emphasis on quantity over quality
  • too much emphasis on safe, easily packaged research
  • too high a pace of publication (as opposed to genuine innovation)
  • difficulty when deciding where to publish interdisciplinary research
  • too much cost (money, environmental impact) of all the conference travel

Solutions:

  • conference papers should not count toward evaluation of an individual researcher
  • professional bodies such as ACM and IEEE should hold fewer conferences (some large and general) and emphasize their journals
  • give more credit for writing reviews, surveys and books
  • use arXiv for all new papers, and create overlay journals

I realise that implementation of these may not be easy, but it has to start somewhere. I wonder whether some people in (T)CS have become addicted to a constant round of conference travel. It really doesn’t seem a good use of society’s resources.

Conferences could retain the role of setting the agenda, selecting the best papers and maintaining a sense of community, through invited plenary lectures, special sessions in trendy areas, etc. Fewer papers would be published, and less repetition would occur. Small invited workshops could still be held, to counterbalance the large open conferences. Reviews and surveys help to set the agenda and allocate attention efficiently, as do prizes and special sessions at meetings.  This system works fine for mathematics. At least for the more mathematical parts of CS, this would work. As for the more “engineering” oriented parts, IEEE also has journals, and requires fast referee reports. It should be possible to improve the CS journal system, if the vast amount of time spent on refereeing for conferences was spent instead on refereeing for journals.