Hector Kohler

I will start a postdoc at École Polytechnique in 2026. During my PhD, I studied Interpretable Reinforcement Learning. Essentially I designed algorithms and benchmarcks to make AI models verifiable, transparent, and with reproducible performances.

Hector Kohler

hector dot kohler at inria.fr

Inria Scool
Université de Lille
Twitter/X
Scholar
Github
Villeneuve d'Ascq
France

Teaching Assistant

Decision under Uncertainty and Reinforcement Learning (Grads, 25 hours).
Sequential Decision Making (Grads, 6 hours).
Algorithms and Programming II (Undergrads, 25 hours).
Algorithms and Data Structures (Undergrads, 36 hours).
Some notebook exercices about Bandits, MDPs, RL, DRL.
Notebooks

Supervisions

Image Segmentation (1 month, U. Lille).
Entropy Regularized RL (2 months, Ecole Polytechnique).
Playing Object Centric Atari with Deep RL (6 months, Ecole Centrale Lille).
Imitating Deep Reinforcement Learning Policies with Decision Trees (6 months, Ecole Centrale Lille).

Open Source

I contribute to rlberry. It is an open source Reinforcement Learning library aimed at education and research. Rlberry main strength is its AgentManager class used to automatise statistically meaningful evaluations of RL agents.
RLberry Github
The API for DPDT is compatible with scikit-learn and available here. Here is an open source code for distillating neural policies in decision trees followin our EWRL17 paper: code.