I am a PhD student in the AI lab of the Vrije Universiteit Brussel (VUB), under the supervision of Ann Nowé (AI Lab, VUB) and Guillermo A. Pérez (Universiteit Antwerpen). My research interests lie in the fields of artificial intelligence and formal verification. More specifically, my PhD focuses on the formal verification of single- and multi-agent policies obtained through reinforcement learning. The end goal of my research is to provide end-users with reliable AI mechanisms.
PhD in Computer Science
Vrije Universiteit Brussel (VUB), Belgium
Master in Computer Science, 2018
University of Mons (UMONS), Belgium
Bachelor in Computer Science, 2016
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.