• DOI:10.1609/aaai.v36i6.20602
• Corpus ID: 245329592

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

@inproceedings{Delgrange2021DistillationOR,
  title={Distillation of RL Policies with Formal Guarantees via Variational Abstraction of Markov Decision Processes (Technical Report)},
  author={Florent Delgrange and Ann Now'e and Guillermo A. P'erez},
  booktitle={AAAI Conference on Artificial Intelligence},
  year={2021}
}

• Florent Delgrange, Ann Now'e, Guillermo A. P'erez
• Published in
  AAAI Conference on Artificial…
  17 December 2021
• Computer Science

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.