A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play
- Submitting institution
-
University College London
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 14566
- Type
- D - Journal article
- DOI
-
10.1126/science.aar6404
- Title of journal
- Science
- Article number
- -
- First page
- 1140
- Volume
- 362
- Issue
- 6419
- ISSN
- 0036-8075
- Open access status
- Exception within 3 months of publication
- Month of publication
- December
- Year of publication
- 2018
- URL
-
-
- Supplementary information
-
https://figshare.com/articles/dataset/_De_Novo_Structure_Prediction_of_Globular_Proteins_Aided_by_Sequence_Variation_Derived_Contacts/964258
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
12
- Research group(s)
-
-
- Citation count
- 306
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- Far-reaching generalisation of AlphaGo Zero, showing that the same algorithm can be used to learn to play perfect-information zero-sum board games at super-human level. Rigorous evaluation against best available game engines in chess, shogi, and Go. Games played by AlphaZero had major impact on the chess community, e.g., book “Game Changer” with game commentaries published by chess masters Matthew Sadler et al. Open source reimplementation Leela Chess Zero won second Top Chess Engine Championship cup 2019.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -