MasterMind double feature: CEC and GECCO

Besides the publication at EvoStar, this year we have managed to publish in the other two big ticket EC conferences of the year, CEC in Cancún and GECCO in Amsterdam. First one on the calendar was CEC, where we presented
A search for scalable evolutionary solutions to the game of MasterMind
as a poster which I tried to make as awesome as possible by using velcro, magnetic strips and whiteboard for a dynamic experience. It
was quite successful indeed, and I expect the message came through well. Here’s the poster, although just the static part of it:


In this poster we attempt to create some guidelines for finding solutions to the game of MasterMind using evolutionary algorithms. We fix the maximum size of the consistent set and use it throughout all problem sizes, although what we discover here is that set size is different for different scoring methods: Entropy or Most Parts. This paper continues the one presented at LION, where we studied only the size needed by Most Parts, which also extended the one presented at GAMEON.

The paper presented at GECCO, which was admitted as full paper, goes in a different direction and is called Improving evolutionary solutions to the game of MasterMind using an entropy-based scoring method. In this case just Entropy is used for scoring (which means that combinations that partition the set of consistent combinations with maximum entropy will be played and sought in the evolutionary search), and the objective was to try and find fast and efficient solutions so that bigger search spaces could be explored. It was pretty much achieved; fastest solution so far (but still slower, mainly due to the nature of Perl, than the solution by Berghman et al.) and best results that those published before by ourselves and others, so all in all we are quite happy about it. In this case I used impress.js for this awesome presentation (which took me quite a while indeed).

EDAs for solving Mastermind at NICSO

The NICSO 2010 talk is taking place as I type this, quite undercrowded due to flight problems, but it’s one of the places where we are presenting new work on Mastermind. In this case, we have presented the paper Adapting Heuristic Mastermind Strategies to Evolutionary Algorithms, by Tom Runarsson and myself. This paper is actually previous to the work we presented in EvoStar. In this paper is where we describe the results we found about the size of the subset of consistent combinations you have to use. OK, I lost you here, but you can watch the presentation

or download it from here. Fee license, as usual, and check the notes for the attribution of pictures.

See ya in PPSN

Today is the last, final and defintive day for submitting papers to the PPSN (Parallel Problem Solving from Nature) conference, and I think we’ve beaten some record here, since our research group (the extended GeNeura, with visiting researchers, other partners in research groups, and so no) is submitting 10 papers. Taking into account the usual PPSN acceptance rate, we’ll probably only manage to get 3 accepted, so anything more than that will really be a success.
There are several evolutionary computation conferences out there, some of them good, some not as good, but for us PPSN is the absolute best. This would be the seudo-20th anniversary (actually, 10th edition, 18th anniversary), and for us it would be our personal 12th anniversary, since the first time we had something admitted in that conference was in 1996, and we haven’t failed a single one (even organized one of them). So even if this year we’re attending all EC conferences there are (CEC, GECCO -yes, GECCO-, Evostar), we really can’t miss this one.
So if you are submitting to PPSN (more than 220, for the time being), good luck, and see you there!