NBIA Colloquium by Søren Hauberg (Technical University of Denmark)

Speaker: Søren Hauberg (Technical University of Denmark)

Title: Invariance in generative Artificial Intelligence

Abstract: Generative AI models learn a compressed representation of data that is often used for downstream tasks such as interpretation, visualization and prediction via transfer learning. Unfortunately, the learned representations are generally not statistically identifiable, leading to a high risk of arbitrariness in the downstream tasks. We propose to use differential geometry to construct representations that are invariant to reparametrizations, thereby solving the bulk of the identifiability problem. We demonstrate that the approach is deeply tied to the uncertainty of the representation, and that practical applications require high-quality uncertainty quantification. With the identifiability problem solved, we show how to construct better priors for generative models, and that the identifiable representations reveals signals in the data that were otherwise hidden.

Brief bio-sketch:  Søren Hauberg is a professor in the Section for Cognitive Systems at the Technical University of Denmark. He received his PhD in computer science from the University of Copenhagen in 2011. Prior to pursuing a PhD he worked as a "digital lumberjack" in the startup Dralle A/S. He was a postdoc for two years at the Max Planck Institute for Intelligent Systems working with Michael Black. In 2013, he was the sole computer science recipient of the Sapere Aude Research Talent award from the Danish Council for Independent Research, and in 2016 he was the sole computer science Villum Young Investigator. In 2017 he was further awarded a Starting Grant from the European Research Council. In 2018, he joined the Young Scientists community under the World Economic Forum, and was in the process named one of "10 of the most exciting young scientists working in the world today." In 2024, he further received the ERC Consolidator grant.

His research interest lie in the span of geometry and statistics. He develops machine learning techniques using geometric constructions, and works on the related numerical challenges. He is particularly interested in random geometries as they naturally appear in learning.