PROJECT TITLE :

Reinforcement Learning in Video Games Using Nearest Neighbor Interpolation and Metric Learning

ABSTRACT:

Reinforcement learning (RL) has had mixed success when applied to games. Large state spaces and also the curse of dimensionality have restricted the power for RL techniques to learn to play advanced games during a affordable length of time. We have a tendency to discuss a modification of Q-learning to use nearest neighbor states to use previous experience in the early stages of learning. A weighting on the state options is learned using metric learning techniques, such that neighboring states represent similar game situations. Our methodology is tested on the arcade game Frogger, and it's shown that some of the effects of the curse of dimensionality can be mitigated.