Saved in:
Bibliographic Details
Main Authors: Bhatia, Abhinav, Nashed, Samer B., Zilberstein, Shlomo
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2306.15909
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912502978183168
author Bhatia, Abhinav
Nashed, Samer B.
Zilberstein, Shlomo
author_facet Bhatia, Abhinav
Nashed, Samer B.
Zilberstein, Shlomo
contents Meta reinforcement learning (Meta-RL) methods such as RL$^2$ have emerged as promising approaches for learning data-efficient RL algorithms tailored to a given task distribution. However, they show poor asymptotic performance and struggle with out-of-distribution tasks because they rely on sequence models, such as recurrent neural networks or transformers, to process experiences rather than summarize them using general-purpose RL components such as value functions. In contrast, traditional RL algorithms are data-inefficient as they do not use domain knowledge, but do converge to an optimal policy in the limit. We propose RL$^3$, a principled hybrid approach that incorporates action-values, learned per task via traditional RL, in the inputs to Meta-RL. We show that RL$^3$ earns a greater cumulative reward in the long term compared to RL$^2$ while drastically reducing meta-training time and generalizes better to out-of-distribution tasks. Experiments are conducted on Meta-RL benchmarks and custom discrete domains that exhibit a range of short-term, long-term, and complex dependencies.
format Preprint
id arxiv_https___arxiv_org_abs_2306_15909
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle RL$^3$: Boosting Meta Reinforcement Learning via RL inside RL$^2$
Bhatia, Abhinav
Nashed, Samer B.
Zilberstein, Shlomo
Machine Learning
Artificial Intelligence
Meta reinforcement learning (Meta-RL) methods such as RL$^2$ have emerged as promising approaches for learning data-efficient RL algorithms tailored to a given task distribution. However, they show poor asymptotic performance and struggle with out-of-distribution tasks because they rely on sequence models, such as recurrent neural networks or transformers, to process experiences rather than summarize them using general-purpose RL components such as value functions. In contrast, traditional RL algorithms are data-inefficient as they do not use domain knowledge, but do converge to an optimal policy in the limit. We propose RL$^3$, a principled hybrid approach that incorporates action-values, learned per task via traditional RL, in the inputs to Meta-RL. We show that RL$^3$ earns a greater cumulative reward in the long term compared to RL$^2$ while drastically reducing meta-training time and generalizes better to out-of-distribution tasks. Experiments are conducted on Meta-RL benchmarks and custom discrete domains that exhibit a range of short-term, long-term, and complex dependencies.
title RL$^3$: Boosting Meta Reinforcement Learning via RL inside RL$^2$
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2306.15909