| Presentation | 2019-03-09 Please fill in Takato Yamazaki, Toshiharu Sugawara, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Multi-Agent Systems (MAS) enable modeling an environment where multiple agents interfere with each other, and it can be applied to many of the real-world problems. However, the environment will be unpredictable and complicated because of the interactions between agents, thereby it is difficult to implement agents’ policy especially when it requires cooperation. In this paper, we address a MAS problem called decentralized multi-task exploration problem, and we observe the emerged agents’ policies which are learned with Deep Q-Network (DQN) implemented in every single agent. In addition, we conduct an experiment in an environment where only limited information is supplied to agents, and we examine how agents will handle the lack of information. As a result, we report that a clear division of labor between agents has occurred in environments where it is difficult to acquire rewards due to information limitation. In that case, each agent's responsible area has little interference from other agents, and it can be handled as an environment similar to a single agent system. Therefore, we experimentally show an improvement of reward acquisition efficiency by using memorized information which can be effectively utilized particularly in a single-agent environment. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Mutli-agent systems / Decentralized systems / Cooperative behavior / Deep reinforcement learning |
| Paper # | AI2018-55 |
| Date of Issue | 2019-03-02 (AI) |
| Conference Information | |
| Committee | AI / IPSJ-ICS / JSAI-KBS / JSAI-DOCMAS / JSAI-SAI |
|---|---|
| Conference Date | 2019/3/7(4days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | Tsunenori Mine(Kyushu Univ.) |
| Vice Chair | Daisuke Katagami(Tokyo Polytechnic Univ.) / Naoki Fukuta(Shizuoka Univ.) |
| Secretary | Daisuke Katagami(Ritsumeikan Univ.) / Naoki Fukuta(Univ. of Electro-Comm.) |
| Assistant | Yuko Sakurai(AIST) |
| Paper Information | |
| Registration To | Technical Committee on Artificial Intelligence and Knowledge-Based Processing / Special Interest Group on Intelligence and Complex Systems / Special Interest Group on Knowledge-Based Systems / Special Interest Group on Data Oriented Constructive Mining and Simulation / Special Interest Group on Society and Artificial Intelligence |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Please fill in |
| Sub Title (in English) | |
| Keyword(1) | Mutli-agent systems |
| Keyword(2) | Decentralized systems |
| Keyword(3) | Cooperative behavior |
| Keyword(4) | Deep reinforcement learning |
| 1st Author's Name | Takato Yamazaki |
| 1st Author's Affiliation | Waseda University(Weseda Univ.) |
| 2nd Author's Name | Toshiharu Sugawara |
| 2nd Author's Affiliation | Waseda University(Weseda Univ.) |
| Date | 2019-03-09 |
| Paper # | AI2018-55 |
| Volume (vol) | vol.118 |
| Number (no) | AI-492 |
| Page | pp.pp.13-18(AI), |
| #Pages | 6 |
| Date of Issue | 2019-03-02 (AI) |