#### DMCA

## Recognizing seki in computer Go (2005)

### Cached

### Download Links

- [www.cs.ualberta.ca]
- [www.fun.ac.jp]
- [webdocs.cs.ualberta.ca]
- [www.is.titech.ac.jp]
- [webdocs.cs.ualberta.ca]
- [webdocs.cs.ualberta.ca]
- DBLP

### Other Repositories/Bibliography

Venue: | IN PROCEEDINGS OF THE 11TH ADVANCES IN COMPUTER GAMES |

Citations: | 2 - 1 self |

### Citations

50 |
Searching with probabilities
- Palay
- 1985
(Show Context)
Citation Context ...e of the unlimited ko threat model, ko is not affected by this. Since df-pn(r) uses a transposition table, in the presence of repetitions the Graph History Interaction (GHI) problem must be addressed =-=[13, 3]-=-. GHI is a notorious problem that may cause search algorithms to incorrectly solve positions. A typical transposition table implementation ignores paths that df-pn(r) takes to cache search results. Ho... |

39 |
Df-pn Algorithm for Searching AND/OR Trees and Its Applications.
- Nagai
- 2002
(Show Context)
Citation Context ... solver is used to prove that stones are safe while territories are not. 5.2 The Df-pn(r) Search Algorithm Df-pn(r) [6, 5] is an extension of Nagai’s depth-first proof-number (df-pn) search algorithm =-=[11]-=-. Df-pn modifies Allis’ best-first proof-number search (PNS) [1] algorithm to use depth-first search. It can expand fewer interior nodes and use a smaller amount of memory than PNS. Df-pn utilizes loc... |

35 |
Computer Go as a sum of local games: An application of combinatorial game theory
- Müller
- 1995
(Show Context)
Citation Context ...ectly. A detailed discussion of the algorithms is provided in the following section. 5 Safety Solver and Algorithms 5.1 Safety Solver The safety solver described in this paper is based on the ones in =-=[8, 12]-=-. It has been integrated into the Go program Explorer [8]. The search-based solver in [12] sequentially processes single regions and tries to prove their safety. To prove a region safe, all boundary b... |

30 |
Life in the game of Go
- Benson
- 1976
(Show Context)
Citation Context ...eft: a static seki. Right: a dynamic seki. Many approaches have been proposed for recognizing safe stones and territories, including both static and search-based methods. Unconditionally alive blocks =-=[2]-=- are safe even when the attacker can play an unlimited number of moves and the defender always passes. Related work on seki analysis also includes Müller’s static analysis and search-based semeai algo... |

20 |
Playing it safe: Recognizing secure territories in computer Go by using static rules and search
- Muller
- 1997
(Show Context)
Citation Context ...ume Go) and safety-of-territory solvers. Müller introduces local search methods for identifying the safety of regions by alternating play, where the defender is allowed to reply to each attacker move =-=[9]-=-. Van der Werf extends Müller’s static rules and uses them in his program to score Go positions [16]. The methods in [9, 12] prove territories safe by using static rules and search. Although some of t... |

19 | The graph-history interaction: On ignoring position history
- Campbell
- 1985
(Show Context)
Citation Context ...e of the unlimited ko threat model, ko is not affected by this. Since df-pn(r) uses a transposition table, in the presence of repetitions the Graph History Interaction (GHI) problem must be addressed =-=[13, 3]-=-. GHI is a notorious problem that may cause search algorithms to incorrectly solve positions. A typical transposition table implementation ignores paths that df-pn(r) takes to cache search results. Ho... |

19 | Static analysis of life and death in the game of Go
- Chen, Chen
- 1999
(Show Context)
Citation Context ...ndamental requirement for programs to always assess the safety of territories correctly. Current computer Go programs use a combination of exact and heuristic techniques, including eye-space analysis =-=[4]-=-, search and heuristic rules based on influence. Heuristic approaches do not guarantee correctness. In order to improve the accuracy of the territory evaluation, they need to be replaced by exact tech... |

19 | Df-pn in Go: Application to the oneeye problem
- Kishimoto, Muller
- 2003
(Show Context)
Citation Context ...cements including iterative deepening, a transposition table, move ordering and heuristic evaluation functions. The safety solver used in this paper utilizes a more powerful df-pn(r) search algorithm =-=[6, 5]-=-. It includes the techniques of [12], such as the solution of strongly and weakly dependent regions, as well as static and heuristic region evaluation functions [12]. The following new features will b... |

17 |
Proof-number search.
- Allis, Meulen, et al.
- 1994
(Show Context)
Citation Context ...re not. 5.2 The Df-pn(r) Search Algorithm Df-pn(r) [6, 5] is an extension of Nagai’s depth-first proof-number (df-pn) search algorithm [11]. Df-pn modifies Allis’ best-first proof-number search (PNS) =-=[1]-=- algorithm to use depth-first search. It can expand fewer interior nodes and use a smaller amount of memory than PNS. Df-pn utilizes local thresholds for both proof and disproof numbers, selects the m... |

17 | A general solution to the graph history interaction problem.
- Kishimoto, Muller
- 2004
(Show Context)
Citation Context ...earch results. However, if a search result that depends on a path is saved in the transposition table, an incorrect cached result may be retrieved from the transposition table. Figure 3, adapted from =-=[7]-=-, shows an example of the GHI problem. Assume that G is a win for the attacker. Let E → H and H → E be pass moves. If H is searched via A → B → E → H, seki is saved in the transposition table entry fo... |

11 | Correct and Efficient Search Algorithms in the Presence of Repetitions
- Kishimoto
- 2005
(Show Context)
Citation Context ...hods that correctly deal with enclosed areas involving seki. The contributions of the paper are summarized as follows: The status ofslocal seki is correctly determined by using the df-pn(r) algorithm =-=[5]-=-. An efficient research method distinguishes a seki from a win for one player by capturing the opponent. An algorithm for solving global seki problems combines the outcome of local seki searches. The ... |

8 |
Race to capture: Analyzing semeai in Go
- Müller
- 1999
(Show Context)
Citation Context ...ted to a semeai, a race to capture between two adjacent groups that cannot both live. For a detailed discussion of semeai, and static methods for evaluation of semeai classes including many seki, see =-=[10]-=-. Figure 1 shows two seki examples. The left Black region is a static seki. Both the defender and the attacker must pass locally, to ensure that their blocks � and � are not captured. They share two l... |

8 |
der Werf. AI Techniques for the Game of Go
- van
- 2004
(Show Context)
Citation Context ...op terminates. All 81 points on this board are proven to be safe for one of the players. 6 Experimental Results Two seki test sets were created by combining examples from several resources, including =-=[14, 17]-=-, and positions from professional games. Test Set 1 is used for local seki testing and Test Set 2 for global seki testing. Both sets contain a mix of easy, moderate and hard problems. They are availab... |

5 | When one eye is sufficient: a static classification
- Vilà, Cazenave
- 2004
(Show Context)
Citation Context ...lso includes Müller’s static analysis and search-based semeai algorithms [10], and Vilà and Cazenave’s static classification rules to recognize safe blocks containing regions up to a size of 7 points =-=[15]-=-, including some seki. Among these static methods, only [10, 15] can detect seki. Search-based methods are used in life and death (tsume Go) and safety-of-territory solvers. Müller introduces local se... |

5 |
Learning to score final positions in the game of Go
- Werf, Herik, et al.
- 2004
(Show Context)
Citation Context ... safety of regions by alternating play, where the defender is allowed to reply to each attacker move [9]. Van der Werf extends Müller’s static rules and uses them in his program to score Go positions =-=[16]-=-. The methods in [9, 12] prove territories safe by using static rules and search. Although some of the static approaches above can handle seki, they do not provide a general solution for recognizing s... |

2 | An improved safety solver for computer Go
- Niu, Müller
- 2006
(Show Context)
Citation Context ...der to improve the accuracy of the territory evaluation, they need to be replaced by exact techniques, using search. An exact, state of the art search-based safety-of-territory solver is described in =-=[12]-=-. However, this solver could not recognize safe stones in seki. The new solver described in this paper extends the previous one and is able to recognize many, even complex seki situations. Seki is a p... |

2 |
1689. Reprinted in Wei Qi Ji Qiao Da
- Pu
- 1996
(Show Context)
Citation Context ...op terminates. All 81 points on this board are proven to be safe for one of the players. 6 Experimental Results Two seki test sets were created by combining examples from several resources, including =-=[14, 17]-=-, and positions from professional games. Test Set 1 is used for local seki testing and Test Set 2 for global seki testing. Both sets contain a mix of easy, moderate and hard problems. They are availab... |