What Drives Compositional Generalization in Visual Generative Models?

Orbis: Overcoming Challenges of Long-Horizon Prediction in Driving World Models

EntityNet: Using Knowledge Graphs to harvest datasets for efficient CLIP model training

Repository containing the code for the Statistical Pattern Recognition exercises.

Official code for the paper "Selective Concept Bottleneck Models Without Predefined Concepts" (TMLR 2025)

Official code for the paper: "When and How Does CLIP Enable Domain and Compositional Generalization?" (ICML 2025 Spotlight)

Official code for the paper "Two Effects, One Trigger: On the Modality Gap, Object Bias, and Information Imbalance in Contrastive Vision-Language Models" (ICLR 2025 Oral)