Mixtures of Experts Unlock Parameter Scaling for Deep RL
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2024
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866914848998162432 |
|---|---|
| author | Obando-Ceron, Johan Sokar, Ghada Willi, Timon Lyle, Clare Farebrother, Jesse Foerster, Jakob Dziugaite, Gintare Karolina Precup, Doina Castro, Pablo Samuel |
| author_facet | Obando-Ceron, Johan Sokar, Ghada Willi, Timon Lyle, Clare Farebrother, Jesse Foerster, Jakob Dziugaite, Gintare Karolina Precup, Doina Castro, Pablo Samuel |
| contents | The recent rapid progress in (self) supervised learning models is in large part predicted by empirical scaling laws: a model's performance scales proportionally to its size. Analogous scaling laws remain elusive for reinforcement learning domains, however, where increasing the parameter count of a model often hurts its final performance. In this paper, we demonstrate that incorporating Mixture-of-Expert (MoE) modules, and in particular Soft MoEs (Puigcerver et al., 2023), into value-based networks results in more parameter-scalable models, evidenced by substantial performance increases across a variety of training regimes and model sizes. This work thus provides strong empirical evidence towards developing scaling laws for reinforcement learning. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2402_08609 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Mixtures of Experts Unlock Parameter Scaling for Deep RL Obando-Ceron, Johan Sokar, Ghada Willi, Timon Lyle, Clare Farebrother, Jesse Foerster, Jakob Dziugaite, Gintare Karolina Precup, Doina Castro, Pablo Samuel Machine Learning Artificial Intelligence The recent rapid progress in (self) supervised learning models is in large part predicted by empirical scaling laws: a model's performance scales proportionally to its size. Analogous scaling laws remain elusive for reinforcement learning domains, however, where increasing the parameter count of a model often hurts its final performance. In this paper, we demonstrate that incorporating Mixture-of-Expert (MoE) modules, and in particular Soft MoEs (Puigcerver et al., 2023), into value-based networks results in more parameter-scalable models, evidenced by substantial performance increases across a variety of training regimes and model sizes. This work thus provides strong empirical evidence towards developing scaling laws for reinforcement learning. |
| title | Mixtures of Experts Unlock Parameter Scaling for Deep RL |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2402.08609 |