On 16 June 2021, Guy Haworth passed way. He was a longtime and active member of our ICCA/ICGA community, and served as a Vice-President of the ICGA from 2002 until 2005. Guy will be most remembered for his contributions to the ICGA Journal. Since 1998, he regularly published technical notes, tournament reports and (review) articles on computer chess. The most well-known contributions were his endgame articles, though over the years he published more and more reports on computer chess tournaments such as the ICGA World Computer Chess Championships, and TCEC (Cup) tournaments. He was an ardent supporter of research on chess endgames, a mentor for many students, and a friend for all who investigated the nucleus of chess.
{"title":"A tribute to Guy Haworth","authors":"H. J. Herik, M. Winands","doi":"10.3233/icg-210190","DOIUrl":"https://doi.org/10.3233/icg-210190","url":null,"abstract":"On 16 June 2021, Guy Haworth passed way. He was a longtime and active member of our ICCA/ICGA community, and served as a Vice-President of the ICGA from 2002 until 2005. Guy will be most remembered for his contributions to the ICGA Journal. Since 1998, he regularly published technical notes, tournament reports and (review) articles on computer chess. The most well-known contributions were his endgame articles, though over the years he published more and more reports on computer chess tournaments such as the ICGA World Computer Chess Championships, and TCEC (Cup) tournaments. He was an ardent supporter of research on chess endgames, a mentor for many students, and a friend for all who investigated the nucleus of chess.","PeriodicalId":14829,"journal":{"name":"J. Int. Comput. Games Assoc.","volume":"5 1","pages":"114-117"},"PeriodicalIF":0.0,"publicationDate":"2021-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84663178","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Unlike AlphaZero-like algorithms (Silver et al., 2018), the Descent framework uses a variant of Unbounded Minimax (Korf and Chickering, 1996), instead of Monte Carlo Tree Search, to construct the partial game tree used to determine the best action to play and to collect data for learning. During training, at each move, the best sequences of moves are iteratively extended until terminal states. During evaluations, the safest action is chosen (after that the best sequences of moves are iteratively extended each until a leaf state is reached). Moreover, it also does not use a policy network, only a value network. The actions therefore do not need to be encoded. Unlike the AlphaZero paradigm, with Descent all data generated during the searches to determine the best actions to play is used for learning. As a result, much more data is generated per game, and thus the training is done more quickly and does not require a (massive) parallelization to give good results (contrary to AlphaZero). It can use end-of-game heuristic evaluation to improve its level of play faster, such as game score or game length (in order to win quickly and lose slowly).
与类似alphazero的算法(Silver et al.， 2018)不同，Descent框架使用Unbounded Minimax (Korf and Chickering, 1996)的变体，而不是Monte Carlo Tree Search，来构建用于确定最佳操作和收集数据以供学习的部分博弈树。在训练过程中，每次移动时，迭代扩展最佳移动序列，直到最终状态。在评估过程中，选择最安全的动作(之后，每次迭代扩展最佳的移动序列，直到达到叶子状态)。此外，它也不使用政策网络，只使用价值网络。因此，不需要对操作进行编码。与AlphaZero范例不同的是，Descent在搜索过程中生成的所有数据都用于学习，以确定最佳操作。因此，每场比赛产生更多的数据，因此训练可以更快地完成，并且不需要(大规模)并行化来获得良好的结果(与AlphaZero相反)。它可以使用游戏结束启发式评估来更快地提高游戏水平，例如游戏分数或游戏长度(为了快速获胜和缓慢失败)。
{"title":"Descent wins five gold medals at the Computer Olympiad","authors":"Quentin Cohen-Solal, T. Cazenave","doi":"10.3233/icg-210192","DOIUrl":"https://doi.org/10.3233/icg-210192","url":null,"abstract":"Unlike AlphaZero-like algorithms (Silver et al., 2018), the Descent framework uses a variant of Unbounded Minimax (Korf and Chickering, 1996), instead of Monte Carlo Tree Search, to construct the partial game tree used to determine the best action to play and to collect data for learning. During training, at each move, the best sequences of moves are iteratively extended until terminal states. During evaluations, the safest action is chosen (after that the best sequences of moves are iteratively extended each until a leaf state is reached). Moreover, it also does not use a policy network, only a value network. The actions therefore do not need to be encoded. Unlike the AlphaZero paradigm, with Descent all data generated during the searches to determine the best actions to play is used for learning. As a result, much more data is generated per game, and thus the training is done more quickly and does not require a (massive) parallelization to give good results (contrary to AlphaZero). It can use end-of-game heuristic evaluation to improve its level of play faster, such as game score or game length (in order to win quickly and lose slowly).","PeriodicalId":14829,"journal":{"name":"J. Int. Comput. Games Assoc.","volume":"60 3","pages":"132-134"},"PeriodicalIF":0.0,"publicationDate":"2021-10-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"72623731","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
PuyoPuyo is one of the Tetris-type games, which is dealt with as a single-player game in this paper. The player has a winning strategy if the player can keep playing the game infinitely on a gameboard of a constant height. In this paper, we consider how lookahead of input pieces affects the existence of winning strategies in PuyoPuyo, and show conditions that the player cannot win even with lookahead. First, we show the number of colors sufficient to make the player lose on the gameboard of width w when the number of lookahead pieces is m. Next, we show that ten and twenty-six colors are sufficient to make the player lose on the gameboards of width two and three, respectively, no matter how large the number of lookahead pieces is.
{"title":"On the power of lookahead in single-player PuyoPuyo","authors":"Yasuhiko Takenaga, Sho Kikuchi, Hushan Quan","doi":"10.3233/icg-210189","DOIUrl":"https://doi.org/10.3233/icg-210189","url":null,"abstract":"PuyoPuyo is one of the Tetris-type games, which is dealt with as a single-player game in this paper. The player has a winning strategy if the player can keep playing the game infinitely on a gameboard of a constant height. In this paper, we consider how lookahead of input pieces affects the existence of winning strategies in PuyoPuyo, and show conditions that the player cannot win even with lookahead. First, we show the number of colors sufficient to make the player lose on the gameboard of width w when the number of lookahead pieces is m. Next, we show that ten and twenty-six colors are sufficient to make the player lose on the gameboards of width two and three, respectively, no matter how large the number of lookahead pieces is.","PeriodicalId":14829,"journal":{"name":"J. Int. Comput. Games Assoc.","volume":"29 1","pages":"102-113"},"PeriodicalIF":0.0,"publicationDate":"2021-10-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"87305746","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The 23rd Computer Olympiad was held November/December 2020. As a consequence of the global pandemic, for the first time the competitions were fought online. This involved creating new rules that were, in large part, dependent on having the participants cooperate on organizing and conducting the events. The event was a success and will be repeated online in 2021.
{"title":"The Computer Olympiad 2020","authors":"H. Iida, J. Schaeffer, I-Chen Wu","doi":"10.3233/icg-210191","DOIUrl":"https://doi.org/10.3233/icg-210191","url":null,"abstract":"The 23rd Computer Olympiad was held November/December 2020. As a consequence of the global pandemic, for the first time the competitions were fought online. This involved creating new rules that were, in large part, dependent on having the participants cooperate on organizing and conducting the events. The event was a success and will be repeated online in 2021.","PeriodicalId":14829,"journal":{"name":"J. Int. Comput. Games Assoc.","volume":"47 1","pages":"118-131"},"PeriodicalIF":0.0,"publicationDate":"2021-10-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"88489502","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
We present a variant of Sudoku called Sudoku Ripeto. It seems to be the first to admit any combination of repeated symbols, and includes Sudoku as a proper subset. We present other Sudoku Ripeto families, each with a different repetition pattern. We define Sudoku Ripeto squares and puzzles, prove several solving rules that generalize those for Sudoku, and give sufficient conditions to flexibly solve puzzles with rules only, without search.
{"title":"Sudoku Ripeto","authors":"Miguel G. Palomo","doi":"10.3233/ICG-210180","DOIUrl":"https://doi.org/10.3233/ICG-210180","url":null,"abstract":"We present a variant of Sudoku called Sudoku Ripeto. It seems to be the first to admit any combination of repeated symbols, and includes Sudoku as a proper subset. We present other Sudoku Ripeto families, each with a different repetition pattern. We define Sudoku Ripeto squares and puzzles, prove several solving rules that generalize those for Sudoku, and give sufficient conditions to flexibly solve puzzles with rules only, without search.","PeriodicalId":14829,"journal":{"name":"J. Int. Comput. Games Assoc.","volume":"94 1","pages":"26-49"},"PeriodicalIF":0.0,"publicationDate":"2021-05-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"76064220","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Dennis J. N. J. Soemers, Vegard Mella, C. Browne, O. Teytaud
Combinations of Monte-Carlo tree search and Deep Neural Networks, trained through self-play, have produced state-of-the-art results for automated game-playing in many board games. The training and search algorithms are not game-specific, but every individual game that these approaches are applied to still requires domain knowledge for the implementation of the game’s rules, and constructing the neural network’s architecture – in particular the shapes of its input and output tensors. Ludii is a general game system that already contains over 1,000 different games, which can rapidly grow thanks to its powerful and user-friendly game description language. Polygames is a framework with training and search algorithms, which has already produced superhuman players for several board games. This paper describes the implementation of a bridge between Ludii and Polygames, which enables Polygames to train and evaluate models for games that are implemented and run through Ludii. We do not require any game-specific domain knowledge anymore, and instead leverage our domain knowledge of the Ludii system and its abstract state and move representations to write functions that can automatically determine the appropriate shapes for input and output tensors for any game implemented in Ludii. We describe experimental results for short training runs in a wide variety of different board games, and discuss several open problems and avenues for future research.
{"title":"Deep Learning for General Game Playing with Ludii and Polygames","authors":"Dennis J. N. J. Soemers, Vegard Mella, C. Browne, O. Teytaud","doi":"10.3233/icg-220197","DOIUrl":"https://doi.org/10.3233/icg-220197","url":null,"abstract":"Combinations of Monte-Carlo tree search and Deep Neural Networks, trained through self-play, have produced state-of-the-art results for automated game-playing in many board games. The training and search algorithms are not game-specific, but every individual game that these approaches are applied to still requires domain knowledge for the implementation of the game’s rules, and constructing the neural network’s architecture – in particular the shapes of its input and output tensors. Ludii is a general game system that already contains over 1,000 different games, which can rapidly grow thanks to its powerful and user-friendly game description language. Polygames is a framework with training and search algorithms, which has already produced superhuman players for several board games. This paper describes the implementation of a bridge between Ludii and Polygames, which enables Polygames to train and evaluate models for games that are implemented and run through Ludii. We do not require any game-specific domain knowledge anymore, and instead leverage our domain knowledge of the Ludii system and its abstract state and move representations to write functions that can automatically determine the appropriate shapes for input and output tensors for any game implemented in Ludii. We describe experimental results for short training runs in a wide variety of different board games, and discuss several open problems and avenues for future research.","PeriodicalId":14829,"journal":{"name":"J. Int. Comput. Games Assoc.","volume":"19 1","pages":"146-161"},"PeriodicalIF":0.0,"publicationDate":"2021-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"74168741","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The TCEC Cup 7 knockout event was the penultimate event of TCEC Season 19, begun on October 31st 2020 with the usual brisk Rapid tempo of 30′+5′′/move. It involved 32 of the top 35 engines of the TCEC19 championship as FIRE, VAJOLET and CHIRON sat this one out, see Fig. 1. It used the rules of TCEC Cup 6 (Haworth and Hernandez, 2020a-2020d). Matches were ‘best of four’ and tie-breaks consisted of further ‘same opening’ mini-matches of two games. For the second time, the ‘equal distance’ pairing was used, with seed s playing seed s+25−r (rather than 26−r−s+1) in round r if the wins all went to the higher seed. Thus, seed s1 plays s17, s9, . . . , s2 if all survive long enough. The higher seed is listed first in Table 1. This pairing also adheres to the Postponement Principle of keeping top seeds apart but stiffens the competition for the top quarter of the seeds and reduces the likelihood of protracting matches far into a tie-break – at least, in the early rounds. Seed s is of course not sentient here and therefore not in a position to wish it was seed s+1.
{"title":"TCEC Cup 7","authors":"G. Haworth, Nelson Hernandez","doi":"10.3233/icg-200174","DOIUrl":"https://doi.org/10.3233/icg-200174","url":null,"abstract":"The TCEC Cup 7 knockout event was the penultimate event of TCEC Season 19, begun on October 31st 2020 with the usual brisk Rapid tempo of 30′+5′′/move. It involved 32 of the top 35 engines of the TCEC19 championship as FIRE, VAJOLET and CHIRON sat this one out, see Fig. 1. It used the rules of TCEC Cup 6 (Haworth and Hernandez, 2020a-2020d). Matches were ‘best of four’ and tie-breaks consisted of further ‘same opening’ mini-matches of two games. For the second time, the ‘equal distance’ pairing was used, with seed s playing seed s+25−r (rather than 26−r−s+1) in round r if the wins all went to the higher seed. Thus, seed s1 plays s17, s9, . . . , s2 if all survive long enough. The higher seed is listed first in Table 1. This pairing also adheres to the Postponement Principle of keeping top seeds apart but stiffens the competition for the top quarter of the seeds and reduces the likelihood of protracting matches far into a tie-break – at least, in the early rounds. Seed s is of course not sentient here and therefore not in a position to wish it was seed s+1.","PeriodicalId":14829,"journal":{"name":"J. Int. Comput. Games Assoc.","volume":"1 1","pages":"321-325"},"PeriodicalIF":0.0,"publicationDate":"2021-01-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89530530","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Season 19’s Premier Division was a gathering of the usual suspects but one participant was not quite what it seemed! STOCKFISH – mighty in Season 18 – had become STOCKFISH NNUE and there was great anticipation of what the added self-learning component to STOCKFISH’s evaluation would mean for STOCKFISH’s strength. In my own engine matches (on much weaker hardware and faster time controls) played from many types of positions, STOCKFISH NNUE had looked extremely impressive against ‘old-fashioned’ STOCKFISH CLASSICAL. Most surprising to me was that STOCKFISH NNUE defended even better than STOCKFISH CLASSICAL which is not a particular strength of self-learning systems. I just wondered whether at long time controls making STOCKFISH ‘think like an NN’ would blunt the destructive power that makes it so formidable. I also had high hopes for ALLIESTEIN which had performed so impressively in the end-of-season-18 bonus competitions.
{"title":"The TCEC19 Computer Chess Superfinal: A perspective","authors":"M. Sadler","doi":"10.3233/icg-200173","DOIUrl":"https://doi.org/10.3233/icg-200173","url":null,"abstract":"Season 19’s Premier Division was a gathering of the usual suspects but one participant was not quite what it seemed! STOCKFISH – mighty in Season 18 – had become STOCKFISH NNUE and there was great anticipation of what the added self-learning component to STOCKFISH’s evaluation would mean for STOCKFISH’s strength. In my own engine matches (on much weaker hardware and faster time controls) played from many types of positions, STOCKFISH NNUE had looked extremely impressive against ‘old-fashioned’ STOCKFISH CLASSICAL. Most surprising to me was that STOCKFISH NNUE defended even better than STOCKFISH CLASSICAL which is not a particular strength of self-learning systems. I just wondered whether at long time controls making STOCKFISH ‘think like an NN’ would blunt the destructive power that makes it so formidable. I also had high hopes for ALLIESTEIN which had performed so impressively in the end-of-season-18 bonus competitions.","PeriodicalId":14829,"journal":{"name":"J. Int. Comput. Games Assoc.","volume":"63 Suppl 1 1","pages":"306-320"},"PeriodicalIF":0.0,"publicationDate":"2021-01-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"88068903","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The last issue of this roller-coaster year contains three scientific articles. The first one, Polygames: Im-proved zero learning , counts 24 authors, which is probably the highest number since the establishment of the journal. The article presents P OLYGAMES , an open-source framework that combines Monte Carlo Tree Search and Deep Learning. The framework is generic enough for implementing many games, being size-invariant, and comes with a games library included. P OLYGAMES won against strong players in the game of 19 × 19 Hex and 8 × 8 Havannah. Fewer authors has the second contri-bution, Analyzing a variant of Clobber: The game of San Jego , by Raphael Thiele and Ingo Althöfer. It introduces a new two-player perfect-information game, called San Jego. The authors establish an upper bound for the state-space complexity and approximate the game-tree complexity. For small board sizes, they calculate the optimal game-theoretic values and investigate the advantage of moving first. Games are fun but can also be used more seriously as the third contribution, A polyomino puzzle for arithmetic practice , by Jeremy Foxcroft and Daniel Ashlock, shows. The article proposes a family of puzzles that gamifies arithmetic skills. The puzzles are designed with an evolutionary algorithm forming an instance of automatic content generation. It a review , con-ference TCEC and
这过山车般的一年的最后一期包含了三篇科学文章。第一篇论文《Polygames: improved -prove zero learning》共有24位作者，这可能是该期刊创刊以来的最高人数。本文介绍了P OLYGAMES，这是一个结合了蒙特卡洛树搜索和深度学习的开源框架。该框架是通用的，足以实现许多游戏，是大小不变的，并附带了一个游戏库。在19 × 19 Hex和8 × 8 Havannah的比赛中，P OLYGAMES战胜了强大的选手。较少的作者有第二个贡献，Raphael Thiele和Ingo Althöfer的《分析clober的变体:the game of San Jego》。它引入了一种新的双人完全信息游戏，叫做San Jego。建立了状态空间复杂度的上界，并对博弈树复杂度进行了近似。对于较小的棋盘，他们计算了最佳的博弈论值，并研究了先动的优势。游戏很有趣，但也可以更严肃地作为第三种贡献，Jeremy Foxcroft和Daniel Ashlock的《算术练习的多利诺谜题》就说明了这一点。这篇文章提出了一系列使算术技能游戏化的谜题。这些谜题是用进化算法设计的，形成了一个自动内容生成的实例。它是一个审查，会议TCEC和
{"title":"Editorial: Many games, many authors","authors":"M. Winands","doi":"10.3233/ICG-200176","DOIUrl":"https://doi.org/10.3233/ICG-200176","url":null,"abstract":"The last issue of this roller-coaster year contains three scientific articles. The first one, Polygames: Im-proved zero learning , counts 24 authors, which is probably the highest number since the establishment of the journal. The article presents P OLYGAMES , an open-source framework that combines Monte Carlo Tree Search and Deep Learning. The framework is generic enough for implementing many games, being size-invariant, and comes with a games library included. P OLYGAMES won against strong players in the game of 19 × 19 Hex and 8 × 8 Havannah. Fewer authors has the second contri-bution, Analyzing a variant of Clobber: The game of San Jego , by Raphael Thiele and Ingo Althöfer. It introduces a new two-player perfect-information game, called San Jego. The authors establish an upper bound for the state-space complexity and approximate the game-tree complexity. For small board sizes, they calculate the optimal game-theoretic values and investigate the advantage of moving first. Games are fun but can also be used more seriously as the third contribution, A polyomino puzzle for arithmetic practice , by Jeremy Foxcroft and Daniel Ashlock, shows. The article proposes a family of puzzles that gamifies arithmetic skills. The puzzles are designed with an evolutionary algorithm forming an instance of automatic content generation. It a review , con-ference TCEC and","PeriodicalId":14829,"journal":{"name":"J. Int. Comput. Games Assoc.","volume":"61 1","pages":"243"},"PeriodicalIF":0.0,"publicationDate":"2021-01-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"76183704","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：