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

Florent Delgrange, Ann Nowe, Guillermo A. Pérez

Onderzoeksoutput: Poster


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 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
StatusUnpublished - 3 jun 2022
EvenementBelgium-Netherlands workshop on Reinforcement Learning 2022 - Leiden University, Leiden, Netherlands
Duur: 3 jun 20223 jun 2022


ConferenceBelgium-Netherlands workshop on Reinforcement Learning 2022
Verkorte titelBeNeRL 2022
Internet adres


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

Citeer dit