[1/3] Excited to share our recent Transactions on Machine Learning Research paper “Semantic Self-adaptation: Enhancing Generalization with a Single Sample” by Sherwin Bahmani, Oliver Hahn, @EduardSebasti20, Nikita Araslanov, Daniel Cremers, and Stefan Roth! Paper: arxiv.org/abs/2208.05788 Code: github.com/visinf/self-ad…

A little late to the party but: 🥳 We are very proud that the students of our lab, together with great collaborators, are going to present 12 amazing works at #CVPR2024! For more info, visit our website: cvg.cit.tum.de

Excited to be in Aistats 🇪🇸🇪🇸🇪🇸 I will present: - Enhancing Hypergradients Estimation: A Study of Preconditioning and Reparameterization arxiv.org/abs/2402.16748 - A Lower Bound and a Near-Optimal Algorithm for Bilevel Empirical Risk Minimization arxiv.org/abs/2302.08766

#CVPR2024 Semantic segmentation with a hierarchical taxonomy: We find that 1) simple baselines ("flat" classifiers) work best; 2) The hyperbolic space mitigates the "parent bias" of the Euclidean space. With Simon Weber, Barış Zöngür and Daniel Cremers.