Prof. Dr. Tobias Huber
Email: | tobias.huber@thithi.de () |
Room: | 2015 (N) |
Address: | Universitätsstraße 6a, 86159 Augsburg |
Links
Research interest
My research focuses on the explainability of Artificial Intelligence and Reinforcement Learning in particular.
I find it fascinating how Reinforcement Learning algorithms can independently develop strategies based only on observations and rewards. In some cases, the agents even develop new strategies that even humans have not yet considered (
e.g. by the chess computer AlphaZero). However, since only the goal of the agents is defined, it is often not clear what exactly the learned strategies look like. This is exacerbated by the use of modern machine learning techniques, which achieve considerable success but are also very opaque.
The goal of my research is to develop new algorithms that make the behaviour of intelligent agents explainable to users and thus facilitate the cooperation between humans and computers.
Talks
2023
An introduction to explanation methods for reinforcement learning and their evaluation. Invited talk in the lecture series "Advanced Topics in AI and Robotics" at the University of Applied Sciences Bonn-Rhein-Sieg, 17.04.2023.
Presenting a demonstrator on Explainable Artificial Intelligence at the opening of the "KI-Erlebnisraums Halle 43". Augsburg, 19.06.2023
2022
Was Pacman denkt – Wie künstliche Intelligenz das Spielen lernt. Talk at the long night of sciences, Augsburg, 16.07.2022, Slides .
2021
Explainable deep Reinforcement Learning. Invited talk in the CSL Machine Learning Reading Club of the Computational Science Lab (CSL) at the University of Hohenheim, 02.02.2021, Slides .
Teaching
Since the winter semester 18/19 I am one of the main lecturers for the courses Game Development and Partical Course Game Development.
Courses
-
WS23/24: Partical Course Game Development
-
SS23 Game Development
-
WS22/23: Partical Course Game Development
-
SS22: Game Development
-
WS21/22: Partical Course Game Development
-
SS21: Game Development
-
WS20/21: Partical Course Game Development
-
SS20: Game Development
-
WS 19/20: Partical Course Game Development
-
SS 19: Game Development
-
WS 18/19: Partical Course Game Development
Insights into final projects
On the courses' website you can see and even play projects from the game programming lectures.
Other lectures I am involved with:
-
WS22/23: Reinforcement Learning
-
WS21/22: Reinforcement Learning
Supervised theses
Master’s Thesis
-
Deep Reinforcement Learning with MuZero: Theoretical Foundations, Variants, and Implementation for a Collaborative Game (2023)
-
Generating Counterfactual Explanations for Atari Agents via Generative Adversarial Networks (2022)
-
Design und Implementierung einer Virtual-Reality-Umgebung zum Testen des impliziten Lernens von motorischen sequentiellen Bewegungen (2021, Link)
- Exploring an Explainable Reinforcement Learning Design for a Self Learning American Football Simulation (2020)
- Implementierung und Vergleich verschiedener Salienz-Karten Algorithmen für tiefes bestärkendes Lernen (2019)
Bachelor’s Thesis
- Explaining the Global Behavior of Deep Reinforcement Learning Agents by Combining T-SNE and Policy Summarization (2023)
- Using Reinforcement Learning to facilitate Implicit Learning in a VR Sports Simulation (2021)
- Understanding Subliminal Persuasive Body Language in Political Speeches via ExplainableArtificial Intelligence (2021)
- Verbinden von Belohnungs-Zerlegung und Strategie-Zusammenfassung für erklärbares Bestärkendes Lernen (2021)
-
Implementation and Comparison of Occlusion-based Explainable Artificial Intelligence Methods (2020, Link)
Editorials & Peer-Review
- AAAI 2024 Workshop on eXplainable AI approaches for deep reinforcement learning (XAI4DRL)
- ECAI 2023: European Conference on Artificial Intelligence - “Quality Champion” award vergeben an 58 von 700 Reviewern mit mindestens 4 Reviews, eines als "Erwartungen übertroffen" bewertet und keines als "unter den Erwartungen".
- KI 2023 – Doctoral Consortium
- AAAI 2022: AAAI Conference on Artificial Intelligence
- IJCAI 2022 Workshop on Explainable Artificial Intelligence (XAI)
I was Reviewer for the following conferences and journals:
Voluntary Activities
Since the end of 2022, I am speaker of the Advisory Board for Young Scientists of the Gesellschaft für Informatik (GI), the largest professional society for computer science in the German-speaking world.
Publications
2024 |
Ruben Schlagwoski, Frederick Herget, Niklas Heimerl, Maximilian Hammerl, Tobias Huber, Pamina Zwolsky, Jan Gruca and Elisabeth André. 2024. From a social POV: the impact of point of view on player behavior, engagement, and experience in a serious social simulation game. In Gillian Smith, Jim Whitehead, Ben Samuel, Katta Spiel, Riemer van Rozen (Eds.). Proceedings of the 19th International Conference on the Foundations of Digital Games, May 21-24, 2024, Worcester, MA, USA. ACM, New York, NY, 34 DOI: 10.1145/3649921.3649936 |
Silvan Mertes, Tobias Huber, Christina Karle, Katharina Weitz, Ruben Schlagowski, Cristina Conati and Elisabeth André. in press. Relevant irrelevance: generating alterfactual explanations for image classifiers. preprint. DOI: 10.48550/arXiv.2405.05295 |
Tobias Huber. 2024. Towards a combined local and global explanation framework for deep reinforcement learning agents with visual input: novel methods and insights from human evaluation. Dissertation, Universität Augsburg. . |
2023 |
Anan Schütt, Tobias Huber, Ilhan Aslan and Elisabeth André. 2023. Fast dynamic difficulty adjustment for intelligent tutoring systems with small datasets. In Mingyu Feng, Tanja Käser, Partha Talukdar (Eds.). Proceedings of the 16th International Conference on Educational Data Mining, 11-14 July 2023, Bengaluru, India. International Educational Data Mining Society, 482-489 |
Tobias Huber, Maximilian Demmler, Silvan Mertes, Matthew Olson and Elisabeth Andrè. 2023. GANterfactual-RL: understanding reinforcement learning agents' strategies through visual counterfactual explanations. In Noa Agmon, Bo An, Alessandro Ricci, William Yeoh (Eds.). AAMAS '23: proceedings of the 2023 International Conference on Autonomous Agents and Multiagent Systems, 29 May - 2 June 2023, London, United Kingdom. International Foundation for Autonomous Agents and Multiagent Systems, Richland, SC, 1097-1106 |
Yael Septon, Tobias Huber, Elisabeth André and Ofra Amir. 2023. Integrating policy summaries with reward decomposition for explaining reinforcement learning agents. Lecture Notes in Computer Science 13955, 320-332. DOI: 10.1007/978-3-031-37616-0_27 |
2022 |
Silvan Mertes, Christina Karle, Tobias Huber, Katharina Weitz, Ruben Schlagowski and Elisabeth André. in press. Alterfactual explanations: the relevance of irrelevance for explaining AI systems. preprint. DOI: 10.48550/arXiv.2207.09374 |
Tobias Huber, Benedikt Limmer and Elisabeth André. 2022. Benchmarking perturbation-based saliency maps for explaining Atari agents. Frontiers in Artificial Intelligence 5, 903875. DOI: 10.3389/frai.2022.903875 |
Pooja Prajod, Dominik Schiller, Tobias Huber and Elisabeth André. 2022. Do deep neural networks forget facial action units? - Exploring the effects of transfer learning in health related facial expression recognition. In Arash Shaban-Nejad, Martin Michalowski and Simone Bianco (Ed.). AI for disease surveillance and pandemic intelligence: intelligent disease detection in action. Springer, Cham (Studies in Computational Intelligence ; 1013), 217-233. DOI: 10.1007/978-3-030-93080-6_16 |
Silvan Mertes, Tobias Huber, Katharina Weitz, Alexander Heimerl and Elisabeth André. 2022. GANterfactual - counterfactual explanations for medical non-experts using generative adversarial learning. Frontiers in Artificial Intelligence 5, 825565. DOI: 10.3389/frai.2022.825565 |
Pooja Prajod, Tobias Huber and Elisabeth André. 2022. Using explainable AI to identify differences between clinical and experimental pain detection models based on facial expressions. Lecture Notes in Computer Science 13141, 311-322. DOI: 10.1007/978-3-030-98358-1_25 |
2021 |
Tobias Huber, Silvan Mertes, Stanislava Rangelova, Simon Flutura and Elisabeth André. 2021. Dynamic difficulty adjustment in virtual reality exergames through experience-driven procedural content generation. In Keeley Crockett, Sanaz Mostaghim, Dipti Srinivasan and Anna Wilbik (Ed.). 2021 IEEE Symposium Series on Computational Intelligence (SSCI), 5-7 December 2021, Orlando, FL, USA. IEEE, Piscataway, NJ, 1-8. DOI: 10.1109/ssci50451.2021.9660086 |
Tobias Huber, Katharina Weitz, Elisabeth André and Ofra Amir. 2021. Local and global explanations of agent behavior: integrating strategy summaries with saliency maps. Artificial Intelligence 301, 103571. DOI: 10.1016/j.artint.2021.103571 |
2020 |
Katharina Weitz, Dominik Schiller, Ruben Schlagowski, Tobias Huber and Elisabeth André. 2020. "Let me explain!": exploring the potential of virtual agents in explainable AI interaction design. Journal on Multimodal User Interfaces 15, 87-98. DOI: 10.1007/s12193-020-00332-0 |
Simon Flutura, Andreas Seiderer, Tobias Huber, Katharina Weitz, Ilhan Aslan, Ruben Schlagowski, Elisabeth André and Joachim Rathmann. 2020. Interactive machine learning and explainability in mobile classification of forest-aesthetics. In Catia Prandi and Johann Marquez-Barja (Ed.). GoodTechs '20: Proceedings of the 6th EAI International Conference on Smart Objects and Technologies for Social Good, September 2020, Antwerp, Belgium. ACM, New York, NY, 90-95. DOI: 10.1145/3411170.3411225 |
Dominik Schiller, Tobias Huber, Michael Dietz and Elisabeth André. 2020. Relevance-based data masking: a model-agnostic transfer learning approach for facial expression recognition. Frontiers in Computer Science 2, 6. DOI: 10.3389/fcomp.2020.00006 |
Klaus Weber, Lukas Tinnes, Tobias Huber, Alexander Heimerl, Eva Pohlen, Marc-Leon Reinecker and Elisabeth André. 2020. Towards demystifying subliminal persuasiveness: using XAI-techniques to highlight persuasive markers of public speeches. Lecture Notes in Computer Science 12175, 113-128. DOI: 10.1007/978-3-030-51924-7_7 |
2019 |
Katharina Weitz, Dominik Schiller, Ruben Schlagowski, Tobias Huber and Elisabeth André. 2019. "Do you trust me?" Increasing user-trust by integrating virtual agents in explainable AI interaction design. In Catherine Pelachaud, Jean-Claude Martin, Hendrik Buschmeier, Gale Lucas and Stefan Kopp (Ed.). Proceedings of the 19th ACM International Conference on Intelligent Virtual Agents - IVA '19, Paris, France, July 02 - 05, 2019. ACM Press, New York, NY, 7-9. DOI: 10.1145/3308532.3329441 |
Tobias Huber, Dominik Schiller and Elisabeth André. 2019. Enhancing explainability of deep reinforcement learning through selective layer-wise relevance propagation. Lecture Notes in Computer Science 11793, 188-202. DOI: 10.1007/978-3-030-30179-8_16 |
Stanislava Rangelova, Simon Flutura, Tobias Huber, Daniel Motus and Elisabeth André. 2019. Exploration of physiological signals using different locomotion techniques in a VR adventure game. Lecture Notes in Computer Science 11572, 601-616. DOI: 10.1007/978-3-030-23560-4_44 |
Dominik Schiller, Tobias Huber, Florian Lingenfelser, Michael Dietz, Andreas Seiderer and Elisabeth André. 2019. Relevance-based feature masking: improving neural network based whale classification through explainable artificial intelligence. In Gernot Kubin, Zdravko Kačič (Eds.). Interspeech 2019, 15-19 September 2019, Graz. ISCA, 2423-2427 DOI: 10.21437/interspeech.2019-2707 |
2018 |
Tobias Huber. 2018. Tiefes bestärkendes Lernen: Grundlagen, Approximationseigenschaft und Implementierung multimodaler Erklärungen. Masterarbeit, Universität Augsburg. Universität Augsburg, Augsburg. |