Critic learning in multi agent credit assignment problem

Article type: Research Article

Authors: Rahaie, Zahra^* | Beigy, Hamid¹

Affiliations: Department of Computer Engineering, Sharif University of Technology, Tehran, Iran

Correspondence: [*] Corresponding author. Zahra Rahaie, Department of Computer Engineering, Sharif University of Technology, Tehran, Iran. Tel.: +9821 6616 6624; Fax: +98 21 66019246; E-mail: z_rahaie@ce.sharif.edu.

Note: [1] E-mail: beigy@sharif.edu.

Abstract: Multi-agent systems can be seen as an apparatus for testing the performance of real distributed systems. One problem encountered in multi-agent systems with the learning capability is credit assignment. This paper presents two methods for solving this problem. The first method assigns credit to the agents according to the history of the interaction while the second method assigns credit to the agents according to the knowledge of agents, and thus the shares of the agents are extracted from the feedback of the environment. The computer experiments show that critic learning has a positive impact in credit assignment problem.

Keywords: Multi-agent systems, credit assignment, reinforcement learning, interaction, history, knowledge

DOI: 10.3233/IFS-162093

Journal: Journal of Intelligent & Fuzzy Systems, vol. 30, no. 6, pp. 3465-3480, 2016