Co-Evolving A Go-Playing Neural Network (2001)
When evolving a game-playing neural network, fitness is usually measured by playing against existing opponents. In this article that need is overcome by applying the competitive co-evolutionary techniques of competitive fitness sharing, shared sampling, and hall of fame to the SANE neuro-evolution method. This approach is tested by evolving networks to play go on small boards. Networks evolved through co-evolution were compared against those evolved against the gnugo program. The co-evolutionary techniques were found to speed up the evolution and to result in better networks not limited by the quality of the opponent.
In Coevolution: {T}urning Adaptive Algorithms Upon Themselves, Birds-of-a-Feather Workshop, Genetic and Evolutionary Computation Conference ({GECCO}-2001), 6, 2001.

Alex Lubberts Former Visitor
Risto Miikkulainen Faculty risto [at] cs utexas edu