B. Bouzy and T. Cazenave "Computer go: An ai-oriented survey," Artificial Intelligence Journal, pp. 39-103, 2001.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....reinforcement: a member calls upon another member when such an action leads to higher reinforcement. 3 Experimental Results 3.1 Experiment Setup One of research in artificial intelligence is programming a computer that can play board games. Board game do mains such as Chess [5] Check [4] GO [3], and Backgammon [17] have been popular since they have finite state spaces with well defined rules. Since it is usually impossible to search exhaustively the state space, artificial intelligence research in game domains has primarily worked on solutions that can play a game comparable to or ....

B. Bouzy and T. Cazenave. Computer go: An ai oriented survey. Artificial Intelligence, 132:39--103, 2001.

....solutions were found for small empty boards up to 5 5, as well as some 6 6 variants. We believe that our system can also be applied to capture, life death and connection problems in the game of Go. 1 Introduction In the last decades, Go has received signi cant attention from AI research [2, 9]. Yet, despite all e orts, the best computer Go programs are still weak. Exemplary is the fact that the largest square board for which a computer proof has been published is only 4 4 [13] According to [15] results for 5 5 and 6 6 exist but are exceedingly subtle and have not been con rmed by ....

....is my move . Enhanced transposition cuto s are used at least three plies from the leaves. 3 The evaluation function The evaluation function is an essential ingredient for guiding the search towards strong play. Unlike in chess, no good and especially no cheap evaluation functions exist for Go [2, 9]. Despite of this we tried to build an evaluation function for the game of Ponnuki Go. The default for solving small games is to use a three valued evaluation function with values [1 (win) 0 (unknown) 1 (loss) Such a threevalued evaluation function is quite ecient for solving games, due to ....

B. Bouzy and T. Cazenave. Computer go: An AI oriented survey. Arti cial Intelligence, 132(1):39-102, October 2001.

....1 Introduction Since the founding years of Arti cial Intelligence (AI) computer games have been used as a test bed for AI algorithms. Many game playing systems have now reached an expert level. Go is a notable exception. In the last decades, Go has received signi cant attention from AI research [3, 13]. Despite all e orts, the best computer Go programs are still weak. Many (if not all) of the current top programs rely on static knowledge bases. As a consequence the programs tend to become extremely complex and dicult to improve. In principle a learning system should be able to overcome this ....

B. Bouzy and T. Cazenave. Computer go: An ai oriented survey. Arti cial Intelligence, 132(1):39-102, October 2001.

No context found.

B. Bouzy and T. Cazenave. Computer go: an ai oriented survey. Artificial Intelligence, 132:39--103, 2001.

....any verification. It costs nothing in execution time but the move generator remains incomplete and always contains errors. When considering the game of go, these two remarks are crucial. Global tree search is 2 not possible in go and knowledge based goprograms are very difficult to improve [Bouzy and Cazenave, 2001]. Therefore, this paper explores an intermediate approach in which a go program performs a global search, not a global tree search, using very little knowledge. This approach is based on statistics or Monte Carlo methods. We believe that such an approach does not have the drawback of go global ....

Bouzy, B. and Cazenave, T. (2001). Computer go: an ai oriented survey. Artificial Intelligence 132, pages 39--103.

....territories, and the score of a given position. In this paper, we shall also use the ( won , undetermined or lost ) terminology when the context is game oriented. This evaluation corresponds to the cognitive model of our thesis [Bouzy 1995a, 1995b] and to the evaluation function description of [Bouzy Cazenave 2001] which gives details of the model used by the quick evaluation function. To give an order of magnitude, the quick evaluation function of a 19x19 board is processed in 50 milliseconds on a 450Mhz computer. This time remains quite stable. It depends on the size of the board and on the number of dead ....

, B. Bouzy, T. Cazenave, "Computer Go : an AI oriented Survey", Artificial Intelligence Journal, Vol 132/1, pp. 39-103, 2001.

....than the complexities of other two person, zero sum, complete information games such as Chess, Shogi, Checkers and Othello. The results achieved by the best go programs are average on the human scale: Wulu, Go4 and Goemate have reached an intermediate level, between a beginner and strong players [2]. The first difficulty to program the game of go lies in the combinatorial complexity which forbids brute force tree search. The second difficulty is to correctly evaluate a position. We have worked on computer go for several years and one of our contributions to computer go was to find the link ....

B. Bouzy, T. Cazenave, "Computer Go : an AI oriented Survey", Artificial Intelligence, Vol. 132 n1 (2001), pp. 39-103.

....times. However, the reduction in search depth provided by the new larger rules decreases much the overall search time, stopping search at smaller depths. 1 Introduction I have written an automatic rule generator based on retrograde analysis of patterns with external conditions [3] 5] [2] . It generates rules about eyes and life in the game of Go. Life and death in the game of go has been already studied, and some clever algorithms have been designed. Beginning with Benson s algorithm [1] that detects unconditional life with the opponent moving as many times he wants to and still ....

B. Bouzy and T. Cazenave, `Computer go: An aioriented survey', Artificial Intelligence, 132(1), 39-- 103, (October 2001).

....measures defined by (7) provide useful information to a Go program or not. This is the dimensional motivation of our paper. 2. 2 the dimensional EF With d integer, we can now define the EF E d by the following formula: E d = S gGb size(g) d S gGw size(g) In our study, we assume that d [0, 2]. E 1 is the classical EF useful for the endgame. E 0 is the count of black groups minus white groups. E 2 measures the ability of one color to get large groups of this color and small groups of the other color. 4 3 Metric evaluation functions This section defines a metric EF. As the ....

....space in which the population evolved was adjusted to the average value for the next period. This adjustment was performed for several reasons. First, to avoid too slow a convergence. For example, if one parameter, say a 2 , converged to a fixed value, say 4, and the population was in the interval [2,6], then generating a program with one parameter set at random in [0,16] was not appropriate. Therefore, a new set of values for this parameter such as [2, 6] with 17 values was chosen. Secondl, when one parameter reached one frontier of the interval, the size of the interval was doubled. For ....

[Article contains additional citation context not shown here]

Bouzy B., Cazenave T., Computer Go : an AI oriented Survey, Artificial Intelligence, Vol. 132 n1 (2001), 39-103

No context found.

B. Bouzy and T. Cazenave "Computer go: An ai-oriented survey," Artificial Intelligence Journal, pp. 39-103, 2001.

No context found.

Bouzy, B., Cazenave, T. (2001) "Computer Go: An AI oriented survey". Artificial Intelligence Vol. 132, number 1, pp. 39-103, 2001.

No context found.

B. Bouzy and T. Cazenave. Computer Go: An AI oriented survey. Artificial Intelligence, 132(1):39--102, October 2001.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC