Bruno Bouzy: Associating Shallow and Selective Global Tree Search with Monte Carlo for 9*9 Go. Computers and Games Bruno Bouzy of Paris Descartes, CPSC, Paris (Paris 5) with expertise in: Artificial Intelligence. Read 73 publications, and contact Bruno Bouzy on ResearchGate. Bruno Bouzy is a player and programmer from France. Born in , his highest rank was 3 dan. He was vice champion of France, losing in the.
|Published (Last):||5 October 2008|
|PDF File Size:||4.27 Mb|
|ePub File Size:||20.65 Mb|
|Price:||Free* [*Free Regsitration Required]|
Moreover, the results of our Monte Carlo programs against knowledge- based programs on 9×9 boards and the ever-increasing power of computers lead us to think that Monte Carlo approaches are worth considering for computer go in the future.
This shows that, on 9×9 boards, Olga is a slightly inferior to Indigo. Skip to search form Skip to main content. It might be possible to link the value of a move to more local subgoals from which we could establish statistics. This paper describes the generation and utilisation of bohzy pattern database for 19×19 go with the Knearest-neighbor representation.
This sole domain- dependent knowledge in Gobble is necessary to ensure that the random games actually finish. Nowadays, some programs are better than human players in most classical games. To support our view, section 2 describes the related work about Monte Carlo applied to go.
The upside of both definitions is the speed of the programs.
Bruno Bouzy – Trích dẫn của Google Scholar
Between each random game, the lists were sorted according to the current evaluation of the moves and then moves were shifted in the list with a probability depending on the temperature. Have you forgotten your login? We set up a match between two versions of Olga using progressive pruning at the root node. The temperature has been implemented in Oleg in a somewhat different way like in Gobble.
The challenge of poker. This progressive pruning algorithm is similar to the one described in [Billings et al. We have set up experiments to assess ideas such as progressive pruning, transpositions, temperature, simulated annealing and depth-two tree search within the Monte Carlo framework.
Finally, it is still slower than classical programs and it is difficult to make them play on boards larger than 13x In the case of a complex domain like the game of Go, … More. Thus, in the beginning, the games were almost completely random, and at the end they were almost completely determined by the evaluations of the moves.
The only domain-dependent knowledge required is the definition of an eye. Then, he evaluated a move by computing the average of the scores of the random games where it had been played.
However, the exact definition of an eye has its importance. On the other … More. Then, section 4 highlights the experiments to validate these ideas. In the latter, two lists of moves were maintained for both players, and the moves in the random games were played in the order of the lists if the move in the list is not legal, we just take the next in the list.
Therefore, the number of candidate moves decreases while the process is running. How do the uses of transpositions and progressive pruning compare in strength? When the tactical module selects moves for the random games, it would be useful for Monte Carlo to use the already available tactical results.
Computer Science > Machine Learning
The re- sult out of 14 9×9 games has been an average 9. Wednesday, April 9, – 1: In this context, each module bouyz independent of the other one, and does not use the strength of the other one.
A review of game-tree pruning. Let us start the random games from the root by two given moves, one move for the friendly side, and, then, one move for the opponent, and make statistics on the terminal position evaluation for each node situated at depth 2 in the min-max tree. We have set up experiments hruno assess ideas … More.
Given the great number of problems and the diversity of possible solutions, Computer Go is an attractive research domain for AI. Conversely, it looks for weaknesses in the opponent position that do not exist.
We would like to know whether and why a move is good. We believe that the statistical search is an alternative to tree search [Junghanns, ]worth considering in practice. Since the beginning of AI, mind games have been studied as relevant application fields.
Inria – Hedging Algorithms and Repeated Matrix Games
Therefore, this paper explores an intermediate approach in which a go program performs bbruno global search, not a global tree search, using very little knowledge. It is based on the same ideas as Gobble; particularly it uses simulated annealing. At least, it provides go programs with a statistical global brubo, which is less expensive than global tree search, and which enriches move generation with a kind of verification.
In such a situation, we use the relevant term of transpositions used within the tree search community [Marsland, ]. Then, it presents our go programs, Olga and Oleg, and it deals with the only important domain-dependent consideration of the method: Thus, brumo obtain a global evaluation of the move. The game of Go is one of the games that still withstand classical Artificial Intelligence approaches.
They are shown in figure 1. In complete information games, the idea of replacing complete information by randomized information is less natural. Game-tree searching by min-max approximation. Inhe tried the expected-outcome model on bouy game of 6×6 Othello. This result underlines that the use of transpositions significantly speeds up the program but decreases its performance. It has already been considered theoretically within the framework of [Rivest, ].