Talk Title:
Vision Language Foundation Models: Tools for Clinical Impact and New Scientific Discovery

Talk Summary:
This lecture explores the fundamental paradigms of modern vision-language foundation models and their similarities/differences to contemporary artificial intelligence (AI) techniques. I will discuss strategies behind the aggregation of large-scale, multimodal data required to train these models and examine their diverse downstream capabilities. Finally, I will demonstrate how foundation models can transcend traditional diagnostic and prognostic tasks to facilitate the discovery of novel scientific hypotheses.

Learning Objectives:

Summarize the major training methodologies for foundation models.
Formulate strategies for evaluating the performance and clinical utility of AI models.
Apply novel AI interpretability approaches to derive and test novel scientific hypotheses.

Presenter Biography:
Dr. Chaudhari is an Associate Professor of Radiology and Biomedical Data Science, and is the Interim Division Chief of the Imaging Informatics research section at Stanford Radiology. Dr. Chaudhari research group develops representation learning and evaluation techniques focusing on large-scale, multi-modal, deep learning models in healthcare that use vision, language, and medical records data. His work has been published in Nature, Nature Medicine, Nature Biomedical Engineering, and numerous machine learning conferences and health journals. Dr. Chaudhari’s research is funded by the NIH, ARPA-H, and several industry partners. He also serves as the Co-Director of Clinical AI at Stanford Radiology and as the Associate Director of Research and Education at the Stanford AIMI Center. He is a co-founder of Cognita Imaging, which received the first FDA breakthrough status for a generative vision language model in radiology.

Author
Akshay Chaudhari
Stanford University