Advanced Topics in AI and Robotics

In this lecture, Huber gives an introduction to the research field of Explainable (Deep) Reinforcement Learning, presents some of the most common methods and shows how their effectiveness can be evaluated both computationally as well as in a user-study using the Arcade Learning environment.

Short Bio

Tobias Huber obtained a Master's degree in Mathematics at the University of Augsburg in 2018. Since then, he is a doctoral student at the University of Augsburg's Chair for Humancentered Artificial Intelligence. Tobias' research aims to facilitate human-AI collaboration by developing novel explainable AI techniques and evaluating their utility for human users. In particular, he focuses on the explainability of Reinforcement Learning agents.

Main papers for the talk

[1] Huber, T., Schiller, D., André, E. (2019). Enhancing Explainability of Deep Reinforcement Learning Through Selective Layer-Wise Relevance Propagation. In: Benzmüller, C., Stuckenschmidt, H. (eds) KI 2019: Advances in Artificial Intelligence. KI 2019. Lecture Notes in Computer Science(), vol 11793. Springer, Cham. https://doi.org/10.1007/978-3-030-30179-8_16

[2] Tobias Huber, Katharina Weitz, Elisabeth André, Ofra Amir, Local and global explanations of agent behavior: Integrating strategy summaries with saliency maps, Artificial Intelligence, Volume 301, 2021.

[3] Huber T, Limmer B, André E. Benchmarking Perturbation-Based Saliency Maps for Explaining Atari Agents. Front Artif Intell. 2022 Jul 13

[4] Huber T, Demmler M, Mertes S, Olson M., André E. GANterfactual-RL: Understanding Reinforcement Learning Agents' Strategies through Visual Counterfactual Explanations. AAMAS 2023

Additional papers, relevant for the Deep RL background

[5] Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., Bellemare, M. G., ... & Hassabis, D. (2015). Human-level control through deep reinforcement learning. nature, 518(7540), 529-533.

The lecture will be held in English and is aimed at students and staff of the H-BRS. Interested parties are cordially invited.