A STUDY ON MODELING AND ANALYSIS OF AGENT-BASED SIMULATIONS WITH Q-LEARNING

摘要

Recently, an agent-based problem has been attracting much attention in various fields, for instance marketing research as well as economic and social sciences. Though many researchers treat this problem, learning effects of agents cannot be disregarded because most agents (persons) often change their action rule according to circumstances. In this paper, we deal with an agent-based problem including Q-learning algorithm, which is one of reinforcement learning methods. In this study, we employ two kinds of agents, followers and pioneers, and design their characters by using Q values. Agents decide their attitudes by ones which neighbor agents show, and then Q values are updated. Simulation studies show that the results of agent-based simulations are affected by action rules of agents and the initial attitudes of agents.