Dr. Christian Gumbsch

I studied Cognitive Science (B.Sc. 2013 and M.Sc. 2018) at the University of Tübingen. Subsequently, I was a research intern and later a doctoral researcher at the Max Planck Institute for Intelligent Systems and the University of Tübingen supervised by Prof. Georg Martius and Prof. Martin Butz. During my doctoral studies I researched how artificial agents can develop temporal abstractions of their sensorimotor experience for hierarchical prediction and planning. My doctoral research followed a highly interdisciplinary approach combining insights from various fields of Cognitive Science, including cognitive and developmental psychology, with deep learning techniques, such as deep reinforcement learning. I joined the Chair of Cognitive and Clinical Neuroscience at the TU Dresden in 2023 to research the encoding of contextual information from multimodal sensory data.

My long-term research goal is to help the development of autonomous agents that learn adaptive goal-directed behavior purely from their interactions with the world, like humans or other animals. To work towards this goal, I follow two research directions that interact bidirectionally: 1. I research inductive biases and computational mechanisms that improve the autonomous learning and problem-solving capabilities of artificial agents. 2. I develop computational cognitive models to research the mechanisms and representations that give rise to the adaptive behavior found in natural agents. Thus, I have a broad research interest in various topic including sequential decision making, hierarchical and multimodal representation learning, and causal discovery in both natural and artificial agents.

Gumbsch, C., Sajid, N., Martius, G., & Butz, M. V. (2024). Learning Hierarchical World Models with Adaptive Temporal Abstractions from Discrete Latent Dynamics. The Twelfth International Conference on Learning Representations (ICLR). https://openreview.net/pdf?id=TjCDNssXKU

Gumbsch, C., Adam, M., Elsner, B., Martius, G., & Butz, M. V. (2022). Developing hierarchical anticipations via neural network-based event segmentation. In 2022 IEEE International Conference on Development and Learning (ICDL), 1-8. IEEE. https://arxiv.org/abs/2206.02042

Eppe, M., Gumbsch, C., Kerzel, M., Nguyen, P. D., Butz, M. V., & Wermter, S. (2022). Intelligent problem-solving as integrated hierarchical reinforcement learning. Nature Machine Intelligence, 4(1), 11-20. https://rdcu.be/cFGsE

Gumbsch, C., Butz, M. V., & Martius, G. (2021). Sparsely changing latent states for prediction and planning in partially observable domains. Advances in Neural Information Processing Systems (NeurIPS), 34, 17518-17531. https://openreview.net/pdf?id=-VjKyYX-PI9

Gumbsch, C., Adam, M., Elsner, B., & Butz, M. V. (2021). Emergent Goal‐Anticipatory Gaze in Infants via Event‐Predictive Learning and Inference. Cognitive Science, 45(8), e13016. https://onlinelibrary.wiley.com/doi/pdfdirect/10.1111/cogs.13016

Gumbsch, C., Butz, M. V., & Martius, G. (2019). Autonomous identification and goal-directed invocation of event-predictive behavioral primitives. IEEE transactions on cognitive and developmental systems, 13(2), 298-311. https://arxiv.org/pdf/1902.09948.pdf