The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
Les méthodes d’apprentissage par renforcement atteignent des performances supérieures à celles des humains dans un large éventail de tâches complexes et d’environnements incertains. Cependant, les hautes performances ne sont pas la seule mesure d’utilisation pratique, comme dans un jeu d’IA ou de conduite autonome. Un agent très efficace agit de manière cupide et égoïste, et est donc peu pratique pour les utilisateurs environnants, d'où une demande pour des agents de type humain. L'apprentissage par imitation reproduit le comportement d'un expert humain et construit un agent semblable à un humain. Cependant, ses performances sont limitées à celles des experts. Dans cette étude, nous proposons un programme de formation pour construire un agent efficace et semblable à un humain en mélangeant l'apprentissage par renforcement et par imitation pour des problèmes d'espace d'action discrets et continus. L'agent hybride proposé atteint des performances supérieures à celles d'un agent d'apprentissage par imitation stricte et présente un comportement plus humain, mesuré via un test de sensibilité humaine.
Rousslan F. J. DOSSA
Kobe University
Xinyu LIAN
Kobe University
Hirokazu NOMOTO
EQUOS RESEARCH Co., Ltd.
Takashi MATSUBARA
Osaka University
Kuniaki UEHARA
Osaka Gakuin University
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copier
Rousslan F. J. DOSSA, Xinyu LIAN, Hirokazu NOMOTO, Takashi MATSUBARA, Kuniaki UEHARA, "Hybrid of Reinforcement and Imitation Learning for Human-Like Agents" in IEICE TRANSACTIONS on Information,
vol. E103-D, no. 9, pp. 1960-1970, September 2020, doi: 10.1587/transinf.2019EDP7298.
Abstract: Reinforcement learning methods achieve performance superior to humans in a wide range of complex tasks and uncertain environments. However, high performance is not the sole metric for practical use such as in a game AI or autonomous driving. A highly efficient agent performs greedily and selfishly, and is thus inconvenient for surrounding users, hence a demand for human-like agents. Imitation learning reproduces the behavior of a human expert and builds a human-like agent. However, its performance is limited to the expert's. In this study, we propose a training scheme to construct a human-like and efficient agent via mixing reinforcement and imitation learning for discrete and continuous action space problems. The proposed hybrid agent achieves a higher performance than a strict imitation learning agent and exhibits more human-like behavior, which is measured via a human sensitivity test.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2019EDP7298/_p
Copier
@ARTICLE{e103-d_9_1960,
author={Rousslan F. J. DOSSA, Xinyu LIAN, Hirokazu NOMOTO, Takashi MATSUBARA, Kuniaki UEHARA, },
journal={IEICE TRANSACTIONS on Information},
title={Hybrid of Reinforcement and Imitation Learning for Human-Like Agents},
year={2020},
volume={E103-D},
number={9},
pages={1960-1970},
abstract={Reinforcement learning methods achieve performance superior to humans in a wide range of complex tasks and uncertain environments. However, high performance is not the sole metric for practical use such as in a game AI or autonomous driving. A highly efficient agent performs greedily and selfishly, and is thus inconvenient for surrounding users, hence a demand for human-like agents. Imitation learning reproduces the behavior of a human expert and builds a human-like agent. However, its performance is limited to the expert's. In this study, we propose a training scheme to construct a human-like and efficient agent via mixing reinforcement and imitation learning for discrete and continuous action space problems. The proposed hybrid agent achieves a higher performance than a strict imitation learning agent and exhibits more human-like behavior, which is measured via a human sensitivity test.},
keywords={},
doi={10.1587/transinf.2019EDP7298},
ISSN={1745-1361},
month={September},}
Copier
TY - JOUR
TI - Hybrid of Reinforcement and Imitation Learning for Human-Like Agents
T2 - IEICE TRANSACTIONS on Information
SP - 1960
EP - 1970
AU - Rousslan F. J. DOSSA
AU - Xinyu LIAN
AU - Hirokazu NOMOTO
AU - Takashi MATSUBARA
AU - Kuniaki UEHARA
PY - 2020
DO - 10.1587/transinf.2019EDP7298
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E103-D
IS - 9
JA - IEICE TRANSACTIONS on Information
Y1 - September 2020
AB - Reinforcement learning methods achieve performance superior to humans in a wide range of complex tasks and uncertain environments. However, high performance is not the sole metric for practical use such as in a game AI or autonomous driving. A highly efficient agent performs greedily and selfishly, and is thus inconvenient for surrounding users, hence a demand for human-like agents. Imitation learning reproduces the behavior of a human expert and builds a human-like agent. However, its performance is limited to the expert's. In this study, we propose a training scheme to construct a human-like and efficient agent via mixing reinforcement and imitation learning for discrete and continuous action space problems. The proposed hybrid agent achieves a higher performance than a strict imitation learning agent and exhibits more human-like behavior, which is measured via a human sensitivity test.
ER -