Convergent and Efficient Deep Q Network Algorithm
This addresses a foundational issue in reinforcement learning by providing a convergent and efficient alternative to DQN, which is incremental but crucial for stability in practical applications.
The paper tackles the problem of DQN's lack of convergence guarantees and instability in realistic settings, proposing a convergent DQN algorithm (C-DQN) that achieves robust learning and solves difficult Atari 2600 games where DQN fails, with support for large discount factors up to 0.9998.
Despite the empirical success of the deep Q network (DQN) reinforcement learning algorithm and its variants, DQN is still not well understood and it does not guarantee convergence. In this work, we show that DQN can indeed diverge and cease to operate in realistic settings. Although there exist gradient-based convergent methods, we show that they actually have inherent problems in learning dynamics which cause them to fail even in simple tasks. To overcome these problems, we propose a convergent DQN algorithm (C-DQN) that is guaranteed to converge and can work with large discount factors (0.9998). It learns robustly in difficult settings and can learn several difficult games in the Atari 2600 benchmark that DQN fails to solve. Our codes have been publicly released and can be used to reproduce our results.