An Evolutionary Algorithm Extended by Ecological Analogy and its Application to the Game of Go

The following two important features of human experts' knowledge are not realized by most evolutionary algorithms: one is that it is various and the other is that the amount of knowledge, including infrequently used knowledge, is large. To imitate these features, we introduce an activation value for individuals and a new variation-making operator, splitting, both of which are inspired by ecological systems. This algorithm is applied to the game of Go and a large amount of knowledge evaluated as appropriate by a human expert is acquired. Various kinds of Go knowledge may be acquired such as patterns, sequences of moves, and Go maxims, part of which has already been realized.