Is Exploration or Optimization the Problem for Deep Reinforcement Learning?
This addresses a foundational issue for researchers in deep RL, though it is incremental as it focuses on diagnosis rather than a new solution.
The paper tackles the problem of distinguishing between exploration and optimization limitations in deep reinforcement learning, finding that deep RL methods exploit only half of the good experience they generate, with a 2-3x gap between best experience and learned performance.
In the era of deep reinforcement learning, making progress is more complex, as the collected experience must be compressed into a deep model for future exploitation and sampling. Many papers have shown that training a deep learning policy under the changing state and action distribution leads to sub-optimal performance, or even collapse. This naturally leads to the concern that even if the community creates improved exploration algorithms or reward objectives, will those improvements fall on the \textit{deaf ears} of optimization difficulties. This work proposes a new \textit{practical} sub-optimality estimator to determine optimization limitations of deep reinforcement learning algorithms. Through experiments across environments and RL algorithms, it is shown that the difference between the best experience generated is 2-3$\times$ better than the policies' learned performance. This large difference indicates that deep RL methods only exploit half of the good experience they generate.