Distributional Bellman Operators over Mean Embeddings
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2023
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866916145599086592 |
|---|---|
| author | Wenliang, Li Kevin Delétang, Grégoire Aitchison, Matthew Hutter, Marcus Ruoss, Anian Gretton, Arthur Rowland, Mark |
| author_facet | Wenliang, Li Kevin Delétang, Grégoire Aitchison, Matthew Hutter, Marcus Ruoss, Anian Gretton, Arthur Rowland, Mark |
| contents | We propose a novel algorithmic framework for distributional reinforcement learning, based on learning finite-dimensional mean embeddings of return distributions. We derive several new algorithms for dynamic programming and temporal-difference learning based on this framework, provide asymptotic convergence theory, and examine the empirical performance of the algorithms on a suite of tabular tasks. Further, we show that this approach can be straightforwardly combined with deep reinforcement learning, and obtain a new deep RL agent that improves over baseline distributional approaches on the Arcade Learning Environment. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2312_07358 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | Distributional Bellman Operators over Mean Embeddings Wenliang, Li Kevin Delétang, Grégoire Aitchison, Matthew Hutter, Marcus Ruoss, Anian Gretton, Arthur Rowland, Mark Machine Learning We propose a novel algorithmic framework for distributional reinforcement learning, based on learning finite-dimensional mean embeddings of return distributions. We derive several new algorithms for dynamic programming and temporal-difference learning based on this framework, provide asymptotic convergence theory, and examine the empirical performance of the algorithms on a suite of tabular tasks. Further, we show that this approach can be straightforwardly combined with deep reinforcement learning, and obtain a new deep RL agent that improves over baseline distributional approaches on the Arcade Learning Environment. |
| title | Distributional Bellman Operators over Mean Embeddings |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2312.07358 |