An integration of individual and social learning in a co-evolutionary approach to the game of Connect-Four

In this paper, we investigate an integration of individual and social learning, utilising co-evolutionary neural networks. Individual learning takes place by playing copies of a player against itself. Social learning allows poor performing players to learn from those players, which are playing at a higher level. The networks are evolved via evolutionary strategies with the network output being used as input to a minimax search tree. Our experiments show that learning is taking place at the 99% confidence level. In terms of performance, the co-evolutionary neural network player has the ability to block two adjacent stones of an opponent.

Text RazaliUTHM.pdf Restricted to Repository staff only Download (60kB)

Actions (login required)

View Item