One-Shot Averaging for Distributed TD($λ$) Under Markov Sampling

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Tian, Haoxing, Paschalidis, Ioannis Ch., Olshevsky, Alex
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913372758343680
author Tian, Haoxing
Paschalidis, Ioannis Ch.
Olshevsky, Alex
author_facet Tian, Haoxing
Paschalidis, Ioannis Ch.
Olshevsky, Alex
contents We consider a distributed setup for reinforcement learning, where each agent has a copy of the same Markov Decision Process but transitions are sampled from the corresponding Markov chain independently by each agent. We show that in this setting, we can achieve a linear speedup for TD($λ$), a family of popular methods for policy evaluation, in the sense that $N$ agents can evaluate a policy $N$ times faster provided the target accuracy is small enough. Notably, this speedup is achieved by ``one shot averaging,'' a procedure where the agents run TD($λ$) with Markov sampling independently and only average their results after the final step. This significantly reduces the amount of communication required to achieve a linear speedup relative to previous work.
format Preprint
id arxiv_https___arxiv_org_abs_2403_08896
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle One-Shot Averaging for Distributed TD($λ$) Under Markov Sampling
Tian, Haoxing
Paschalidis, Ioannis Ch.
Olshevsky, Alex
Machine Learning
Distributed, Parallel, and Cluster Computing
We consider a distributed setup for reinforcement learning, where each agent has a copy of the same Markov Decision Process but transitions are sampled from the corresponding Markov chain independently by each agent. We show that in this setting, we can achieve a linear speedup for TD($λ$), a family of popular methods for policy evaluation, in the sense that $N$ agents can evaluate a policy $N$ times faster provided the target accuracy is small enough. Notably, this speedup is achieved by ``one shot averaging,'' a procedure where the agents run TD($λ$) with Markov sampling independently and only average their results after the final step. This significantly reduces the amount of communication required to achieve a linear speedup relative to previous work.
title One-Shot Averaging for Distributed TD($λ$) Under Markov Sampling
topic Machine Learning
Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2403.08896