Abstract | ||
---|---|---|
A key challenge in model-based reinforcement learning (RL) is to synthesize computationally efficient and accurate environment models. We show that carefully designed generative models that learn and operate on compact state representations, so-called state-space models, substantially reduce the computational costs for predicting outcomes of sequences of actions. Extensive experiments establish that state-space models accurately capture the dynamics of Atari games from the Arcade Learning Environment from raw pixels. The computational speed-up of state-space models while maintaining high accuracy makes their application in RL feasible: We demonstrate that agents which query these models for decision making outperform strong model-free baselines on the game MSPACMAN, demonstrating the potential of using learned environment models for planning. |
Year | Venue | Field |
---|---|---|
2018 | arXiv: Learning | Artificial intelligence,Learning environment,Pixel,Generative grammar,Machine learning,Mathematics,Reinforcement learning |
DocType | Volume | Citations |
Journal | abs/1802.03006 | 12 |
PageRank | References | Authors |
0.46 | 20 | 11 |
Name | Order | Citations | PageRank |
---|---|---|---|
Lars Buesing | 1 | 248 | 16.50 |
Theophane Weber | 2 | 159 | 16.79 |
Sébastien Racanière | 3 | 28 | 1.42 |
s m ali eslami | 4 | 119 | 8.58 |
Danilo Jimenez Rezende | 5 | 1567 | 81.67 |
David P. Reichert | 6 | 88 | 6.85 |
Fabio Viola | 7 | 202 | 8.87 |
Frederic Besse | 8 | 100 | 5.17 |
Karol Gregor | 9 | 1173 | 72.53 |
Demis Hassabis | 10 | 4924 | 191.12 |
Daan Wierstra | 11 | 5412 | 255.92 |