My research focuses on multi-modal machine learning (e.g., vision and language) and applications to health and broad science. Our recent works explain the representation space geometry from multi-modal contrastive learning (ICLR'24, ICLR'23, NeurIPS'22), investigate emergent behaviors of multi-modal foundation models (ICLR'23, ACL'23, NeurIPS'23, TMLR'23), solve novel tasks by compounding vision-language foundation models (CVPR'24, Preprint'24, Preprint'23), and adapt these foundation models to health (ML4H'22, JAMIA'21, npj Digital Medicine'19, npj Digital Medicine'18). I have also contributed to many open-source systems, including Stanza, HELM, Jiuge, and OpenLM (ACL'20, TMLR'22, ACL'19).