Solving Games with Perfect Information using Genetic Algorithms and a Backpropagation Neural Network

R.F. Bouroncle Cuba (Peru)


Games with perfect information, Genetic algorithms, Backpropagation net


The problem with complex games is that there are many ways to play them. In other words the group of possible strategies is large. However, if we consider a strategy as a succession of pure strategies through generations in the game (as a game tree), we can solve the problem with a Genetic Algorithm (GA). Furthermore, fitness function which evaluates the individuals needs to know the state of the game, in order to apply the best strategy. Representing discreetly the state of the game (pattern) we can train a Backpropagation Net (BPN) to recognize it. This research work proposes a method to solve games with perfect information from Artificial Intelligence's perspective. The use of this method will be shown on the development of a hybrid system, which can play a Chess departure successfully. The advantage between this method and the extensive method is that the number of individuals in each generation is always the same (only the best strategies survive).

Important Links:

Go Back