Reinforcement Learning on Atari 57 Human Starts
567Mean Human-Normalized ScorePrior. Duel Clip
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Prior. Duel ClipArchitecture=Dueling, Experience Replay=Prioritized, Gradient Clipping=true2015.11 | 567 | 115.3 | |
| Prior. SingleArchitecture=Single, Experience Replay=Prioritized, Gradient Clipping=true2015.11 | 386.7 | 112.9 | |
| Duel ClipArchitecture=Dueling, Experience Replay=Uniform, Gradient Clipping=true2015.11 | 343.8 | 117.1 | |
| SingleArchitecture=Single, Experience Replay=Uniform, Gradient Clipping=false2015.11 | 332.9 | 110.9 | |
| Single ClipArchitecture=Single, Experience Replay=Uniform, Gradient Clipping=true2015.11 | 302.8 | 114.1 | |
| Nature DQNArchitecture=Nature DQN2015.11 | 219.6 | 68.5 |