samedi 8 décembre 2012

The danger of unbalanced PC


CPAIOR is one of my favourite conferences because it “provides an opportunity for researchers in one area to learn about techniques in the others”.
A few years ago, Jean-Charles complained that there was no PC member from ILOG (now IBM) at CP2008. I agree with him it was not enough considering the role this company plays in the field.
Last year at CPAIOR12, 7% of the CP members were from IBM.
This year it’s 22%. I’m not sure people from IBM will complain any more ;-)

Although I have nothing against this company, I feel a bit uncomfortable with this percentage and I can see some risk there. Let me illustrate this with a tweet from LocalSolver (not IBM) (commercial local search solver):

“Don't publish dedicated heuristic approaches without having tried ‪#LocalSolver on your combinatorial problem: your paper might be rejected!”

I don’t know if this was a joke or not, but to me it illustrates perfectly the danger there is if a same affiliation is over-represented in a PC of a conference. There are other important players (commercial or open-source) also building nice solvers that are not represented at all in this PC committee… (for instance localsolver ;-))

I don’t blame anybody and I like the fact the people from industry are represented in PC. This is only because I like CPAIOR that I make this remark. But I’m sure I’m not the only one looking at the PC members composition before considering submitting a paper and it would be really sad if people don’t submit because they feel uncomfortable with the PC composition.

What do you think should be the maximum percentage of people with a same affiliation in such a conference? Maybe this should be written in the status of the conference?


8 commentaires:

  1. This post is misleading as readers may think the tweet you cite is from IBM.

    RépondreSupprimer
  2. I've put the "a tweet from LocalSolver (not IBM)" in bold and red. I hope you find it less misleading.

    RépondreSupprimer
  3. Thank you.

    You use some arguably unethical tweet from an IBM competitor to illustrate a yet to be found danger from IBM. I still don't see the connection here.

    Note that I would be worried to see the person who wrote that tweet in a programm committee as her judgement would be clearly biased.

    RépondreSupprimer
  4. Jean François, this an extract of review I received this year:

    "The authors compare their CP approach with a commercial MIP solver but
    not a commercial CP solver. It would be good to have a comparison to
    see how much their method improves over what is available in CP currently."

    I let you guess what commercial CP solver the reviewer means here ?

    RépondreSupprimer
  5. Jean-François,
    The reference to the tweet was effectively misleading. However, the question asked by Pierre remains a good questions.
    We can have different opinions, but, for me, 22% is too much.
    The advantage is that they all will come to the conference in order to show that they were really motivated to be involved in the conference (and not only in the PC) :-)

    RépondreSupprimer
  6. @pschaus I don't have a crystal ball. I guess (but have no evidence) that the comment is from an OR person who is used to see comparisons with state of the art solvers in OR, ie comparison with one of the top commercial solvers. He or she assumed the same should be true on the CP side. I don't read this as an IBM conspiracy. (and I didn't write that comment either -) )

    @regin I am arguing with the use of misleading information, not with your opinion. I respect your opinion and that of Pierre.

    RépondreSupprimer
  7. That's interesting, what do you think is a state of the art CP solver?
    The only testing roughly equivalent to Mittleman benchmarks (on which MIP solvers always compare themselves) is the minizinc challenge. According to last results I should rather compare to Gecode.

    RépondreSupprimer
  8. Actually, considering the last minizinc challenge results, we should all compare to Chuffed which seems far more efficient than Gecode ...

    RépondreSupprimer