Deep reinforcement learning algorithms require large amounts of experience to
learn an individual task. While in principle meta-reinforcement learning
(meta-RL) algorithms enable agents to learn new skills from small amounts of
experience, several major challenges preclude their practicality. Current
methods rely heavily on on-policy experience, limiting their sample efficiency.
The also lack mechanisms to reason about task uncertainty when adapting to new
tasks, limiting their effectiveness in sparse reward problems. In this paper,
we address these challenges by developing an off-policy meta-RL algorithm that
disentangles task inference and control. In our approach, we perform online
probabilistic filtering of latent task variables to infer how to solve a new
task from small amounts of experience. This probabilistic interpretation
enables posterior sampling for structured and efficient exploration. We
demonstrate how to integrate these task variables with off-policy RL algorithms
to achieve both meta-training and adaptation efficiency. Our method outperforms
prior algorithms in sample efficiency by 20-100X as well as in asymptotic
performance on several meta-RL benchmarks.
Description
[1903.08254] Efficient Off-Policy Meta-Reinforcement Learning via Probabilistic Context Variables
%0 Journal Article
%1 rakelly2019efficient
%A Rakelly, Kate
%A Zhou, Aurick
%A Quillen, Deirdre
%A Finn, Chelsea
%A Levine, Sergey
%D 2019
%K meta-learning reinforcement-learning
%T Efficient Off-Policy Meta-Reinforcement Learning via Probabilistic
Context Variables
%U http://arxiv.org/abs/1903.08254
%X Deep reinforcement learning algorithms require large amounts of experience to
learn an individual task. While in principle meta-reinforcement learning
(meta-RL) algorithms enable agents to learn new skills from small amounts of
experience, several major challenges preclude their practicality. Current
methods rely heavily on on-policy experience, limiting their sample efficiency.
The also lack mechanisms to reason about task uncertainty when adapting to new
tasks, limiting their effectiveness in sparse reward problems. In this paper,
we address these challenges by developing an off-policy meta-RL algorithm that
disentangles task inference and control. In our approach, we perform online
probabilistic filtering of latent task variables to infer how to solve a new
task from small amounts of experience. This probabilistic interpretation
enables posterior sampling for structured and efficient exploration. We
demonstrate how to integrate these task variables with off-policy RL algorithms
to achieve both meta-training and adaptation efficiency. Our method outperforms
prior algorithms in sample efficiency by 20-100X as well as in asymptotic
performance on several meta-RL benchmarks.
@article{rakelly2019efficient,
abstract = {Deep reinforcement learning algorithms require large amounts of experience to
learn an individual task. While in principle meta-reinforcement learning
(meta-RL) algorithms enable agents to learn new skills from small amounts of
experience, several major challenges preclude their practicality. Current
methods rely heavily on on-policy experience, limiting their sample efficiency.
The also lack mechanisms to reason about task uncertainty when adapting to new
tasks, limiting their effectiveness in sparse reward problems. In this paper,
we address these challenges by developing an off-policy meta-RL algorithm that
disentangles task inference and control. In our approach, we perform online
probabilistic filtering of latent task variables to infer how to solve a new
task from small amounts of experience. This probabilistic interpretation
enables posterior sampling for structured and efficient exploration. We
demonstrate how to integrate these task variables with off-policy RL algorithms
to achieve both meta-training and adaptation efficiency. Our method outperforms
prior algorithms in sample efficiency by 20-100X as well as in asymptotic
performance on several meta-RL benchmarks.},
added-at = {2019-04-20T20:32:53.000+0200},
author = {Rakelly, Kate and Zhou, Aurick and Quillen, Deirdre and Finn, Chelsea and Levine, Sergey},
biburl = {https://www.bibsonomy.org/bibtex/225c891cc27443eb25c3cc4392b2c7284/kirk86},
description = {[1903.08254] Efficient Off-Policy Meta-Reinforcement Learning via Probabilistic Context Variables},
interhash = {19101e4e1177e83491c880e5a1463461},
intrahash = {25c891cc27443eb25c3cc4392b2c7284},
keywords = {meta-learning reinforcement-learning},
note = {cite arxiv:1903.08254},
timestamp = {2019-04-20T20:32:53.000+0200},
title = {Efficient Off-Policy Meta-Reinforcement Learning via Probabilistic
Context Variables},
url = {http://arxiv.org/abs/1903.08254},
year = 2019
}