Output details

Learning mazes with aliasing states: An LCS algorithm with associative perception

<24>This paper is the culmination of a project that explored employing ideas in psychological learning theory in a reinforcement learning scenario. The essential idea is to adapt the concepts of imprinting and the laws of organisation to learn to differentiate indeterminate environments based on delayed stimulus. We propose the most extensive test suite of maze problems available, define several new operators for learning classifiers and conduct the most rigorous evaluation of learning classifier system on maze problems to date. We were able to learn strategies for solving aliasing mazes at least as quickly and using less memory than other approaches.