Multi-hop Learning Promote Cooperation in Multi-agent Systems

Abstract

The behavior of individuals maximizing their own benefits in some systems leads to decline of system performance. A challenge remains to promote and maintain cooperation between selfish individuals in multi-agent systems. We propose a multi-hop learning method to promote cooperation in multi-agent system. Based on spatial evolutionary dilemma game, agents can learn strategies from multi-hop neighbors on grid network to improve overall system’s fitness. We investigate the system’s cooperation rate with different hop learning ability in the Prisoner’s Dilemma game, the Snowdrift game and the Stag-hunt game. Experiments show that for Stag-hunt game and Prisoner’s Dilemma game, multi-hop learning is a fairly good way to promote cooperation in multi-agent systems.