Are you conducting research in Artificial Intelligence and considering applying for an AI Junior Research Group or an ERC Starting Grant?

The Campus Institute Data Science (CIDAS) at the University of Göttingen, in collaboration with various departments, invites you to a digital information event. The event will provide insights into funding opportunities, support services, and the research environment in Göttingen.

On May 8th, 2025, we welcome Prof. Michel Besserve from the Technical University of Braunschweig to the CIDAS Colloquium.

The talk will take place in Room 2.101, followed by a get-together in a nearby room on the same floor.

The presentation, titled “Functional and causal inductive biases for deep generative model representations”, explores how deep generative models can learn meaningful internal representations of the external world.

Prof. Besserve discusses the importance of inductive biases—both functional and causal—for ensuring identifiability in latent variable models. He highlights assumptions on the model's function space and the causal structure of the data-generating process.

The talk also illustrates how these biases are implicitly used in methods like variational autoencoders and self-supervised learning, offering insight into their empirical success. Finally, ongoing work on causal model reduction will be presented, aiming to map complex high-dimensional models to simpler, interpretable ones.