Two Improved Algorithms Based on DQN

DQN (Deep Q-Network) is a deep reinforcement learning algorithm, proposed by DeepMind researchers in 2013, which is a value function-based reinforcement learning method designed to solve reinforcement learning problems with high-dimensional state spaces and discrete action spaces. So far, many research teams have made improvements to the DQN algorithm, in this paper, on the basis of four existing improved algorithms, two groups of them are selected for integration, which we call NDP_DQN and NDP51_DQN, respectively. We have conducted experiments with the DQN as a baseline and the two improved algorithms as a comparison on Atari 2600, an important experimental platform in the field of Deep Reinforcement Learning, and the experiments show that both algorithms exhibit higher performance over the original DQN algorithm on an arcade game in the Atari 2600.