A Boosting-based Deep Neural Networks Algorithm for Reinforcement Learning

Yu Wang, Hongxia Jin

Abstract: In this paper, a new boosting-based deep neural networks algorithm is designed for improving the performance of model-free reinforcement learning structures. Based on theoretical proof and performance analysis, it is going to demonstrate that the new approach gives a faster convergence speed and a better return compared to existing deep neural network based RL approaches, like deep <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$Q$</tex> network (DQN)[30], [13] and etc. A complete and detailed exploration of the new algorithm will be given in the paper as well. Also, simulation studies are conducted and compared with several other RL algorithms on the RL benchmark experiment tasks as given in [22]. The results demonstrate a great performance improvement on <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$Q$</tex> learning by using our new boosting-based deep neural networks algorithm.