Video Understanding for Science

Title: Video Understanding for Science

Bio: Jennifer is an assistant professor joining Cornell CIS this fall. Her research focuses on developing computer vision & machine learning systems for scientists to scalably analyze increasing amounts of experimental data. She aims to accelerate scientific discovery and optimize expert attention in real-world workflows, tackling challenges including annotation efficiency, model interpretability and generalization, and semantic structure discovery. 

Abstract:  Scientists are capturing increasing amounts of video data across fields such as ethology, neuroscience, and ecology. Analyzing these videos at scale requires automated systems that work together with scientists to extract meaningful insights, as downstream tasks are diverse and require expert interpretation. Achieving these goals demands close collaboration between AI researchers and domain experts. This talk presents the development of video analysis systems for science, focusing on animal behavior analysis in behavioral neuroscience. We will discuss the potential of new developments, such as video foundation models, and their ability to perform diverse tasks – classification, localization, retrieval – from a single, frozen model with minimal adaptation. We aim to develop collaborative systems that accelerate discovery and expand our understanding of the natural world.