Cross-Validation in Multiagent-based Simulation: Analyzing Evolutionary Bargaining Agents
Keiki Takadama, Yutaka L. Suematsu, Norberto Eiji Nawa, Katsunori Shimohara · Tokyo Tech Research Repository (Tokyo Institute of Technology) · 2006
This paper addresses cross-validation in multiagent-based simulation by analyzing evolutionary agents in a bargaining game in game theory. In particular, this paper focuses on analyzing different learning mechanisms and knowledge representation capabilities applied to agents for cross-validation. To investgate hese issues, we compare the following two cases: (1) agents employing an evolutionary stratgy (ES) and agents employing a learnig classifier system (LCS) as different learning mechanisms; and (2) agents handling an ordinary explanation of numbers and agents handling a limited explanation of numbers as different knowlege representation capabilities. An intensive comparison of simulation results reveal the following implications: (1) simulation results b ES-based agents show the same tendency in game theory but those by LCS-based agents do not; and (2) even simulation results by ES-based agents become strange when the agents are restricted to handling only a real number with two decimal dgits instead of an ordinary real number in a negotiation process between the agents in the bagaining game.