English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

We consider the challenge of policy simplification and verification in thecontext of policies learned through reinforcement learning (RL) in continuousenvironments. In well-behaved settings, RL algorithms have convergenceguarantees in the limit. While these guarantees are valuable, they areinsufficient for safety-critical applications. Furthermore, they are lost whenapplying advanced techniques such as deep-RL. To recover guarantees whenapplying advanced RL algorithms to more complex environments with (i)reachability, (ii) safety-constrained reachability, or (iii) discounted-rewardobjectives, we build upon the DeepMDP framework introduced by Gelada et al. toderive new bisimulation bounds between the unknown environment and a learneddiscrete latent model of it. Our bisimulation bounds enable the application offormal methods for Markov decision processes. Finally, we show how one can usea policy obtained via state-of-the-art RL to efficiently train a variationalautoencoder that yields a discrete latent model with provably approximatelycorrect bisimulation guarantees. Additionally, we obtain a distilled version ofthe policy for the latent model.

Distillation of RL Policies with Formal Guarantees via Variational Abstraction of Markov Decision Processes (Technical Report)