Henry's research focuses on using theoretically-grounded machine learning to train, understand, and reliably leverage foundation models.