论文信息 - An Acceleration Method for Evolutionary Systems Based on Iterated Prisoner's Dilemma

An Acceleration Method for Evolutionary Systems Based on Iterated Prisoner's Dilemma

In this paper, we describe an acceleration method for evolutionary systems based on Iterated Prisoner's Dilemma (IPD). In the systems, agents play IPD games with other agents, and strategies which earn more payoffs will gradually increase their ratio in the total population. Most computation time of the systems is occupied by IPD games, which are repeatedly played among the agents. In our method, repetition of moves in IPD games are detected, and payoffs by those moves are calculated at once by multipliers. In order to maximize the number of units for IPD games on one FPGA, multipliers are shared by the agents. We have implemented a system with the method on XC2VP100, and the speedup by the method is about four times compared with a system without the method.

[1] Tsutomu Maruyama,et al. High Speed Hardware Computation of Co-evolution Models , 1999, ECAL.

[2] K. Lindgren,et al. Cooperation and community structure in artificial ecosystems , 1993 .

[3] Tsutomu Maruyama,et al. A Co-processor System with a Virtex FPGA for Evolutionary Computation , 2000, FPL.

[4] Kristian Lindgren,et al. Evolutionary phenomena in simple dynamics , 1992 .

[5] J. Neumann,et al. Prisoner's Dilemma , 1993 .