Jean-Marc Alliot

From Chessprogramming wiki
Jump to: navigation, search

Home * People * Jean-Marc Alliot

Jean-Marc Alliot,
a French mathematician, computer scientist and head of the optimization and high performance computing department at Institut de Recherche en Informatique de Toulouse (IRIT, Toulouse Computer Science Research Institute), which is a joint research unit of Toulouse Universities and the National Center for Scientific Research (CNRS). He received his Ph.D. in computer science in 1992 from Paul Sabatier University, Toulouse III on implementing Prolog extensions of a parallel inference engine under supervision of Luis Fariñas del Cerro, and habilitated in operations research and mathematical programming in 1996 at National Polytechnic Institute of Toulouse under Joseph Noailles on the topic of aircraft conflict resolution using Genetic Algorithms (GA). GA was also topic in his joined effort along with Nicolas Durand to improve an Othello program [2]. His research interests further includes a broad range of artificial intelligence, artificial evolution, information theory, mathematical optimization, temporal logic and bioinformatics.

Who is the Master?

As a chess lover [3], Jean-Marc Alliot proposed a novel approach based on a Markovian interpretation of the game that would rank the greatest chess masters more fairly than the Elo system [4]. In his study, elaborated and published in the April 2017 ICGA Journal under the title Who is the Master? [5], 26,000 games (over 2 million positions) played at regular time control by all world champions since Wilhelm Steinitz have been analyzed using Stockfish 190915 [6], running on a cluster of 640 AMD 6262 HE Opteron processors in 62000 CPU hours with multiPV 2 and 4GB hash for each instance, in order to create Markov matrices for each year a player was active based on the conformance of his moves. For each position, the model estimates the probability of making a mistake, and the magnitude of the mistake by comparing the two best moves calculated at an average of 2 minutes by move (26 plies on average) with the move actually played, starting from move number 10 [7]. By using classical linear algebra methods on these matrices, the outcome of games between any players can be predicted, and this prediction is shown to be at least as good as the Elo prediction for players who actually played each other.

See also

Selected Publications


1992 ...

2000 ...

2010 ...

Forum Discussions

External Links