论文信息 - Kernel-Based Online NEAT for Keepaway Soccer

Kernel-Based Online NEAT for Keepaway Soccer

This paper presents a kernel-based online neuroevolutionary of augmenting topology (KO-NEAT) algorithm, which borrowing the selection mechanisms used in temporal difference (TD) algorithms and combining the kernel function approximator for individual fitness initiation. KO-NEAT can improve evolution's online performance of NEAT and learns more quickly. Empirical results in keepaway soccer problem demonstrate that KO-NEAT can substantially improve the original algorithm.

[1] Peter Stone,et al. Layered learning in multiagent systems - a winning approach to robotic soccer , 2000, Intelligent robotics and autonomous agents.

[2] Shimon Whiteson,et al. On-line evolutionary computation for reinforcement learning in stochastic domains , 2006, GECCO.

[3] Shimon Whiteson,et al. Comparing evolutionary and temporal difference methods in a reinforcement learning domain , 2006, GECCO.

[4] Risto Miikkulainen,et al. Evolving Neural Networks through Augmenting Topologies , 2002, Evolutionary Computation.

[5] Peter Stone,et al. Keepaway Soccer: A Machine Learning Testbed , 2001, RoboCup.

[6] Peter Stone,et al. Keepaway Soccer: From Machine Learning Testbed to Benchmark , 2005, RoboCup.

[7] Zhang Ru-bo,et al. Reinforcement Learning Theory,Algorithms and Its Application , 2000 .

[8] Lu Ji-lian. Overview of Robotsoccer Learning Method , 2004 .

[9] Risto Miikkulainen,et al. Competitive Coevolution through Evolutionary Complexification , 2011, J. Artif. Intell. Res..

[10] Risto Miikkulainen,et al. Evolving a Roving Eye for Go , 2004, GECCO.

[11] Richard S. Sutton,et al. Learning to predict by the methods of temporal differences , 1988, Machine Learning.

[12] Andrew W. Moore,et al. Reinforcement Learning: A Survey , 1996, J. Artif. Intell. Res..