Enregistré dans:
| Auteurs principaux: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2024
|
| Sujets: | |
| Accès en ligne: | https://arxiv.org/abs/2405.03967 |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866911869940269056 |
|---|---|
| author | Gogineni, Kailash Dayapule, Sai Santosh Gómez-Luna, Juan Gogineni, Karthikeya Wei, Peng Lan, Tian Sadrosadati, Mohammad Mutlu, Onur Venkataramani, Guru |
| author_facet | Gogineni, Kailash Dayapule, Sai Santosh Gómez-Luna, Juan Gogineni, Karthikeya Wei, Peng Lan, Tian Sadrosadati, Mohammad Mutlu, Onur Venkataramani, Guru |
| contents | Reinforcement Learning (RL) trains agents to learn optimal behavior by maximizing reward signals from experience datasets. However, RL training often faces memory limitations, leading to execution latencies and prolonged training times. To overcome this, SwiftRL explores Processing-In-Memory (PIM) architectures to accelerate RL workloads. We achieve near-linear performance scaling by implementing RL algorithms like Tabular Q-learning and SARSA on UPMEM PIM systems and optimizing for hardware. Our experiments on OpenAI GYM environments using UPMEM hardware demonstrate superior performance compared to CPU and GPU implementations. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2405_03967 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | SwiftRL: Towards Efficient Reinforcement Learning on Real Processing-In-Memory Systems Gogineni, Kailash Dayapule, Sai Santosh Gómez-Luna, Juan Gogineni, Karthikeya Wei, Peng Lan, Tian Sadrosadati, Mohammad Mutlu, Onur Venkataramani, Guru Machine Learning Artificial Intelligence Hardware Architecture Reinforcement Learning (RL) trains agents to learn optimal behavior by maximizing reward signals from experience datasets. However, RL training often faces memory limitations, leading to execution latencies and prolonged training times. To overcome this, SwiftRL explores Processing-In-Memory (PIM) architectures to accelerate RL workloads. We achieve near-linear performance scaling by implementing RL algorithms like Tabular Q-learning and SARSA on UPMEM PIM systems and optimizing for hardware. Our experiments on OpenAI GYM environments using UPMEM hardware demonstrate superior performance compared to CPU and GPU implementations. |
| title | SwiftRL: Towards Efficient Reinforcement Learning on Real Processing-In-Memory Systems |
| topic | Machine Learning Artificial Intelligence Hardware Architecture |
| url | https://arxiv.org/abs/2405.03967 |