Reinforcement Learning in Game Playing

Abstract

This study is conducted to understand the internal workings of reinforcement learning. In the movie called "Terminator II", in a clip, Arnold Schwarzeneger told the little boy he was protecting from the Terminator that "My CPU is a neural net computer. The more I interact with humans, the more I will learn and understand about humans." Reinforcement learning (RL) is one mechanism that improves an agent's intelligence by evaluating the feedback that it receives from the environment with which it interacts. RL rewards well chosen actions and punishes bad decisions. The RL algorithm that was experimented with in this study is Q-Iearning. The agent was given the task of learning to play the trivial game of tic-tac-toe. Without any winning strategy encoded into the agent, the agent improved its moves selection by playing against its opponent. The first half of the study examined th parameters of Q-learning. The second half of this study used a neural network to generalize the agent's experience. The advantages and disadvantages of both generalized Q-Iearning and Q-learning are discussed.