An investigation on piece differential information in co-evolution on games using Kalah

This paper describes a series of experiments using co-evolution of artificial neural networks on a game called Kalah. The technique employed closely follows the one used by Chellapilla and Fogel to evolve the successful checkers program Anaconda. The experiments aim to provide insight on the effect of including piece differential information, a basic yet crucial piece of expert knowledge, into the neural network inputs.