The objective of this research was to investigate the possibility of using Neural Networks to create artificial pool players that solve the game in the same way that humans would. In the base of the learning algorithm is positive and negative reinforcement on a back-propagation learning neural network. The study showed what the most appropriate format for representing the table is and how to create and interpret the output of the network. It highlights the main problems of the system such as foul-turns and oscillations into local minima and gives solutions to counteract them.
You can view a demo of the simulaton and network playing only after 3 games: http://youtu.be/lcjwg13xUQk