Prioritized Experience Replay
This improves reinforcement learning efficiency for agents in complex environments like Atari games, though it is incremental as it builds on the existing DQN framework.
The paper tackled the problem of inefficient learning in reinforcement learning by prioritizing important experiences for replay, resulting in a new state-of-the-art performance where DQN with prioritized replay outperformed uniform replay on 41 out of 49 Atari games.
Experience replay lets online reinforcement learning agents remember and reuse experiences from the past. In prior work, experience transitions were uniformly sampled from a replay memory. However, this approach simply replays transitions at the same frequency that they were originally experienced, regardless of their significance. In this paper we develop a framework for prioritizing experience, so as to replay important transitions more frequently, and therefore learn more efficiently. We use prioritized experience replay in Deep Q-Networks (DQN), a reinforcement learning algorithm that achieved human-level performance across many Atari games. DQN with prioritized experience replay achieves a new state-of-the-art, outperforming DQN with uniform replay on 41 out of 49 games.