Distillation of RL Policies with Formal Guarantees via Variational Abstraction of Markov Decision Processes

Florent Delgrange, Ann Nowé, Guillermo A. Pérez

Onderzoeksoutput: Conference paper

15 Downloads (Pure)

Samenvatting

We consider the challenge of policy simplification and verification in the context of policies learned through reinforcement learning (RL) in continuous environments. In well-behaved settings, RL algorithms have convergence guarantees in the limit. While these guarantees are valuable, they are insufficient for safety-critical applications. Furthermore, they are lost when applying advanced techniques such as deep-RL. To recover guarantees when applying advanced RL algorithms to more complex environments with (i) reachability, (ii) safety-constrained reachability, or (iii) discounted-reward objectives, we build upon the DeepMDP framework introduced by Gelada et al. to derive new bisimulation bounds between the unknown environment and a learned discrete latent model of it. Our bisimulation bounds enable the application of formal methods for Markov decision processes. Finally, we show how one can use a policy obtained via state-of-the-art RL to efficiently train a variational autoencoder that yields a discrete latent model with provably approximately correct bisimulation guarantees. Additionally, we obtain a distilled version of the policy for the latent model.
Originele taal-2English
TitelProceedings of the AAAI Conference on Artificial Intelligence
SubtitelVol. 36 No. 6: AAAI-22 Technical Tracks 6
Plaats van productiePalo Alto, California USA
UitgeverijAAAI Press
Hoofdstuk6
Pagina's6497-6505
Aantal pagina's9
Volume36
UitgaveFirst
ISBN van geprinte versie1-57735-876-7, 978-1-57735-876-3
DOI's
StatusPublished - 28 jun 2022
Evenement36th AAAI Conference on Artificial Intelligence -
Duur: 22 feb 20221 mrt 2022
Congresnummer: 36
https://aaai.org/Conferences/AAAI-22/

Publicatie series

NaamProceedings of the AAAI Conference on Artificial Intelligence
UitgeverijAAAI Press
Nummer6
Volume36
ISSN van geprinte versie2159-5399
ISSN van elektronische versie2374-3468

Conference

Conference36th AAAI Conference on Artificial Intelligence
Verkorte titelAAAI
Periode22/02/221/03/22
Internet adres

Vingerafdruk

Duik in de onderzoeksthema's van 'Distillation of RL Policies with Formal Guarantees via Variational Abstraction of Markov Decision Processes'. Samen vormen ze een unieke vingerafdruk.

Citeer dit