Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Liang, Anthony, Tennenholtz, Guy, Hsu, Chih-wei, Chow, Yinlam, Bıyık, Erdem, Boutilier, Craig
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2402.15957
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866912143858728960
author Liang, Anthony
Tennenholtz, Guy
Hsu, Chih-wei
Chow, Yinlam
Bıyık, Erdem
Boutilier, Craig
author_facet Liang, Anthony
Tennenholtz, Guy
Hsu, Chih-wei
Chow, Yinlam
Bıyık, Erdem
Boutilier, Craig
contents We introduce DynaMITE-RL, a meta-reinforcement learning (meta-RL) approach to approximate inference in environments where the latent state evolves at varying rates. We model episode sessions - parts of the episode where the latent state is fixed - and propose three key modifications to existing meta-RL methods: consistency of latent information within sessions, session masking, and prior latent conditioning. We demonstrate the importance of these modifications in various domains, ranging from discrete Gridworld environments to continuous-control and simulated robot assistive tasks, demonstrating that DynaMITE-RL significantly outperforms state-of-the-art baselines in sample efficiency and inference returns.
format Preprint
id arxiv_https___arxiv_org_abs_2402_15957
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle DynaMITE-RL: A Dynamic Model for Improved Temporal Meta-Reinforcement Learning
Liang, Anthony
Tennenholtz, Guy
Hsu, Chih-wei
Chow, Yinlam
Bıyık, Erdem
Boutilier, Craig
Machine Learning
We introduce DynaMITE-RL, a meta-reinforcement learning (meta-RL) approach to approximate inference in environments where the latent state evolves at varying rates. We model episode sessions - parts of the episode where the latent state is fixed - and propose three key modifications to existing meta-RL methods: consistency of latent information within sessions, session masking, and prior latent conditioning. We demonstrate the importance of these modifications in various domains, ranging from discrete Gridworld environments to continuous-control and simulated robot assistive tasks, demonstrating that DynaMITE-RL significantly outperforms state-of-the-art baselines in sample efficiency and inference returns.
title DynaMITE-RL: A Dynamic Model for Improved Temporal Meta-Reinforcement Learning
topic Machine Learning
url https://arxiv.org/abs/2402.15957