Hilti Win276






Multi-Layers 3D-Adaptive Model for Iterated N-Players Prisoners' Dilemma Based on Particle Swarm Optimization

Multi-Layers 3D-Adaptive Model for Iterated N-Players Prisoners' Dilemma Based on Particle Swarm Optimization
1500 until 1600
Meeting Room 7th Floor
Sally H. Al-Manasra
Associate Professor Muhammad Rafie Hj. Mohd. Arshad
Game Theory is a mathematical tool that can analyze interactions between individuals strategically. The interactions between several agents, who may be individuals, groups or firms, are interdependent. The interdependent interactions are controlled by the available strategies and payoff of participants. However, the strategies nature of social interaction makes game theory useful to employ in different applications. Since any interaction between two parties has an outcome, the outcome of the interaction of any individual player depends on its own choices and on the choices made by other individuals in the corresponding game.
Game Theory had successfully analyzed the Prisoner’s Dilemma as the best known example of social Dilemmas. Prisoner’s dilemma can be either one-shot game or iterated game. In this research we are interested in the iterated n-player prisoner’s dilemma (INPPD) since this version of game is applicable to many real-life applications.
Finding the optimal strategy in a given game with large number of players is a complex problem. The related problem is to find out the strategy that a large group of rational opponents follow throughout the game (opponent modeling). In addition, the problem of existing optimization models is that these models support limited number of players. For that reason, in this research we constructed an adaptive 3D model that is efficient in optimization INPPD problem with the assistant of Particles Swarm Optimization (PSO). The model consists of several components including a set of adaptive automata, dynamic knowledge-base and an intelligent searching component that is based on PSO.
The analysis showed that our model could improve the performance of participated players significantly throughout the generations of particles in the swarm. These improvements include increasing the number of cooperated agents, increasing the cooperated actions made by the players in both sub-swarm and swarm levels, increasing the payoffs achieved by the best players in all sub-swarms. Finally, the model could successfully achieve efficient results on well-known benchmarks.