Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

Rainbow: Combining Improvements in Deep Reinforcement Learning

About

The deep reinforcement learning community has made several independent improvements to the DQN algorithm. However, it is unclear which of these extensions are complementary and can be fruitfully combined. This paper examines six extensions to the DQN algorithm and empirically studies their combination. Our experiments show that the combination provides state-of-the-art performance on the Atari 2600 benchmark, both in terms of data efficiency and final performance. We also provide results from a detailed ablation study that shows the contribution of each component to overall performance.

Matteo Hessel, Joseph Modayil, Hado van Hasselt, Tom Schaul, Georg Ostrovski, Will Dabney, Dan Horgan, Bilal Piot, Mohammad Azar, David Silver• 2017

Related benchmarks

TaskDatasetResultRank
Reinforcement LearningAtari 2600 MONTEZUMA'S REVENGE
Score154
45
Atari Game PlayingPitfall!
Score0.00e+0
25
Reinforcement LearningAtari100k (test)
Alien Score318.7
23
Reinforcement LearningAtari 2600 57 games
Median Human-Normalized Score223
20
Reinforcement LearningALE Atari 57 games
HWRB4
16
Reinforcement LearningAtari 2600 57 games (test)
Median Human-Normalized Score231.1
15
Deadline Compliance Scheduling200 Randomized Tasksets u ≈ 0.87 1.0
Mean Compliance Rate78
15
Sudoku SolvingSudoku 2x2
Final Reward1.3
14
Atari Game PlayingAtari 2600 57 games human starts evaluation metric
Median Human-Normalized Score153
14
Reinforcement LearningAtari-57 (full)
HWRB4
13
Showing 10 of 42 rows

Other info

Follow for update