CS Special Seminar: Azade Farshad "From Generative Visual Models towards Counterfactual World Models"

From Generative Visual Models towards Counterfactual World Models

Azade Farshad
Technical University of Munich
Google Scholar

Abstract: This talk explores the limitations of current foundation models, which, despite their impressive capabilities with multimodal data and scale, often lack crucial domain-specific knowledge. I will present my research trajectory in computer vision and healthcare, demonstrating how incorporating semantic understanding significantly enhances the reliability of the models for image and video analysis. Building on these insights, I will outline my vision for next-generation AI systems that move beyond data-scaling approaches by systematically integrating structured knowledge, specifically, causal reasoning frameworks informed by physical, chemical, and physiological principles. This research direction addresses fundamental challenges in modeling complex interactions and their effects, enabling more robust prediction and interpretation across domains.

Bio: Dr. Azade Farshad is a postdoctoral research associate at the Computer Science department of the Technical University of Munich, Germany, where she leads the Medical Image Analysis and GenAI research group at the CAMP Chair. She is also affiliated with the Munich Center for Machine Learning and is a visiting scholar at the STAI lab at Stanford. Her core research investigates the frontiers of image and video generation and manipulation, alongside foundational scene analysis, with applications spanning both general computer vision and specialized healthcare domains. Dr. Farshad has demonstrated significant leadership in the field, serving as the primary organizer of the SG2RL workshop at ICCV 2023 and CVPR 2024, and as a co-organizer for the MELEX and WiCV workshops at ECCV 2024 and CVPR 2024. Her expertise is further recognized through her roles as Associate Program Chair at the IPMI 2023 conference and Area Chair for MICCAI 2025. Actively contributing to the research community, she is an Executive Committee member of the British Machine Vision Association and a board member of Women in MICCAI. Dr. Farshad's exceptional contributions have been acknowledged with the prestigious best dissertation award at the German Conference on Medical Image Computing (BVM) 2025. Her innovative research is widely recognized through publications in leading international conferences, including CVPR, ICCV, MICCAI, and NeurIPS.