2018 Fiscal Year Final Research Report
Machine learning and distributed game-tree search in games
| Project/Area Number |
16H02927
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (B)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Entertainment and game informatics 1
|
|---|
| Research Institution | The University of Tokyo |
|---|
Principal Investigator |
KANEKO TOMOYUKI 東京大学, 大学院情報学環・学際情報学府, 准教授 (00345068)
|
|---|
| Project Period (FY) |
2016-04-01 – 2019-03-31
|
| Keywords | ゲームプログラミング |
|---|
| Outline of Final Research Achievements |
After the success of AlphaGo, many researchers are focusing on games to develop technologies in artificial intelligence. In this research project, we have developed new techniques in machine learning (reinforcement learning) and game tree search to improve the performance of AI agents on games. Our research involves parallel and/or distributed computing to speed-up learning, because the learning requires tremendous amount of computing resources in the existing techniques in reinforcement learning.
|
| Free Research Field |
ゲーム情報学
|
| Academic Significance and Societal Importance of the Research Achievements |
研究計画の通りに機械学習とゲーム木探索とその並列実行について研究した．また採択後に大きく発展したAlphaGoに代表するGPU計算と強化学習についても，最新技術を研究に反映させた．成果の一つであるUniformity regularizationという学習方式については，tensorflow及びchainer上で実装し，囲碁，将棋，チェスなど代表的なゲームを題材に性能を示した．成果全体を総合して15件の論文を公表した．
|
