Implementation of reinforcement learning using neural network

Abstract

Reinforcement learning is an area of machine learning solving the problems that how to take actions to get optimal goals in some certain environment. One kind of reinforcement learning algorithm—Q-learning integrated with neural network is proposed in this project to improve the performance of reinforcement learning algorithm. This paper will present the implementation of the Q-learning with backpropagation neural network. The programming algorithm and its functions are discussed in details. The performance of the algorithms and its influencing factors are tested in the mountain car problem benchmark. The results indicate that reinforcement learning using neural network is feasible and outperform with mass of data. A summary of the project and future work will also be provided in the end.