SwiftRL: Towards Efficient Reinforcement Learning on Real Processing-In-Memory Systems

arXiv:2405.03967v1 Announce Type: new
Abstract: 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 …

abstract agents architectures arxiv behavior cs.ar cs.lg datasets experience however in-memory learn limitations linear memory near processing reinforcement reinforcement learning systems training trains type workloads