Title :
Multi-agent emergent self-organization using an appropriate reward function
Author :
Hercog, Luis Miramontes
Author_Institution :
Monterrey Technol. Inst., Mexico City, Mexico
Abstract :
This paper shows the importance of the reward function and distribution in a game theoretical framework, the game at hand is the "El Farol" Bar Problem well known for the need of inductive reasoning to find the solution. The agents are evolutionary learners which perception is very simple. The first results show no adaptation using the traditional reward function of the Minority Game. Then, a new reward function which is layered and positive in all its domain is introduced. Using the new reward function, the multiagent system adapts to the problem through emergent behavior, appearing an agent that is changing side all the time, called the vacillating agent. The agent balances the system allowing some agents to fix on both sides, the bar and at home, producing a Nash equilibrium, hence the optimal performance of the system.
Keywords :
evolutionary computation; game theory; inference mechanisms; learning (artificial intelligence); multi-agent systems; pattern classification; El Farol Bar Problem; Minority Game; Nash equilibrium; game theoretical framework; inductive reasoning; multiagent system; reward function; vacillating agent; Appropriate technology; Artificial intelligence; Drives; Evolutionary computation; Game theory; Intelligent systems; Iron; Learning systems; Mathematical model; Nash equilibrium;
Conference_Titel :
Cybernetics and Intelligent Systems, 2004 IEEE Conference on
Print_ISBN :
0-7803-8643-4
DOI :
10.1109/ICCIS.2004.1460383