Keynote Talk: Ehsan Adeli

Dealing with Data Scarcity and Imperfections Using Multi-Site Data and Longitudinal Self-Supervised Representation Learning


In this talk, I will focus on two techniques we recently developed to deal with data scarcity and imperfections. First, I will discuss how we can learn good representations from MRIs using the self-supervised signal obtained by exploiting the repeated measures in longitudinal studies. Second, I will present our series of works for correcting data/feature distribution with respect to extraneous or confounding metadata variables. I will explain how the operation could be used to harmonize the data from multiple sites with different characteristics to build a unified learning framework.

Speaker’s Bio

Dr. Ehsan Adeli is an Assistant Professor at the Department of Psychiatry and Behavioral Sciences. He is also affiliated with the Department of Computer Science. He primarily leads research at the Computational Neuroscience (CNS) Lab, and the Stanford AI Lab (SAIL), Stanford Vision and Learning (SVL) lab, and the Partnership in AI-Assisted Care (PAC). His research interests include computer vision, computational neuroscience, medical image analysis, and AI-assisted healthcare. Dr. Adeli is an Executive Co-Director of Stanford AGILE Consortium (Advancing technoloGy for fraIlty & LongEvity), and a faculty member of Stanford Wu Tsai Neurosciences Institute, Stanford Institute for Human-Centered AI, and Stanford Center for AI in Medical Imaging. He is an Associate Editor of IEEE Journal of Biomedical and Health Informatics and the Journal of Ambient Intelligence and Smart Environments. He is a Senior Member of IEEE and has recently served as area chair or associate editor for several top conferences.