KataGo
| Original author(s) | David Wu |
|---|---|
| Developer(s) | David Wu |
| Initial release | 27 February 2019 |
| Stable release | 1.10.0
/ 24 October 2021 |
| Engine | |
| Type | Go software |
| License | MIT License |
| Website | katagotraining |
Search KataGo on Amazon.
KataGo is a free and open-source computer Go program released on 27 February 2019. It is developed by David Wu[1] of Jane Street Capital.
Like DeepMind's AlphaGo Zero, KataGo uses Monte Carlo tree search with a convolutional neural network providing position evaluation and policy guidance. KataGo introduces many refinements that enable it to learn faster and play more strongly.[1][2] Notable features of KataGo that are absent in many other Go-playing programs include score and final-point-ownership estimation (so KataGo will predict not only that White will win with probability 80%, but that White will win by 3 points and which points will be White's and which Black's); the ability to play strongly with boards of different sizes, different values of komi, and in handicap games; and the ability to use a variety of different rulesets and adjust its play and evaluation for the small differences between them.
KataGo's first release was trained by David Wu using resources provided by his employer, but it is now trained by a distributed effort, coordinated at the website https://katagotraining.org/. Members of the community provide computing resources by running the client, which generates self-play games and rating games submits them to the server. The self-play games are used to train newer networks and the rating games to evaluate the networks' relative strength.
KataGo is widely used by strong human go players, including the South Korean national team, for training purposes[3][4].
A specially trained version of KataGo was used to make progress[5] on the formidably difficult and still not definitively solved 120th problem in the Igo Hatsuyōron. Starting with what was at the time the strongest version of the KataGo neural network, training was conducted for a week using positions chosen randomly from variations beginning at the problem's position. Over the course of the week, KataGo appears to have discovered many of the same difficult ideas as humans exploring the problem, as well as new moves not previously considered by humans.
References
- ↑ 1.0 1.1 "Accelerating Self-Play Learning in Go". arxiv.org. 27 February 2019. Retrieved 13 October 2021.
- ↑ David Wu (15 November 2020). "Other Methods Implemented in KataGo".
- ↑ 金雷 (1 March 2021). "AI当道 中国围棋优势缩小了吗？" [With the dominance of AI, is China's Go superiority shrinking?]. Xinmin Evening News. Retrieved 5 December 2021.
- ↑ Hong-ryeol Lee (14 April 2020). "'AI 기사' 격전장에 괴물 '블랙홀'이 등장했다" [A monster 'black hole' appeared in the battlefield of 'AI knight']. The Chosun Ilbo. Retrieved 8 December 2021.
- ↑ "A presentation of the techniques used and the main results obtained". Retrieved 6 December 2021.
External links
Category:Go engines Category:Free and open-source software Category:2019 software Category:Applied machine learning
This article "KataGo" is from Wikipedia. The list of its authors can be seen in its historical and/or the page Edithistory:KataGo. Articles copied from Draft Namespace on Wikipedia could be seen on the Draft Namespace of Wikipedia and not main one.
|
This page exists already on Wikipedia. |
