Yi Li, Ph.D.

Yi Li is a CTML Postdoctoral Scholar in the Division of Biostatistics at UC Berkeley School of Public Health, mentored by Mark van der Laan. Yi's research interests span targeted learning, causal inference, and survival analysis, with a growing interest in agentic AI and AI for science.

Yi's dissertation, "Regularized Targeted Learning under Analytic Intractability," developed methods for stable estimation and reliable inference when the usual tools of targeted learning, such as a tractable efficient influence curve (the formula guiding the targeting step) or a variance formula, are unavailable. His dissertation introduced regularized TMLE (targeted maximum likelihood estimation) in models implied by the highly adaptive lasso, a flexible machine learning method; Targeted Deep Architectures, which embed TMLE inside a neural network; and a grouped V-fold jackknife, a resampling method for standard errors without a variance formula. These ideas were applied to survival data with two partially observed event times.

At CTML, Yi works within the Novo Nordisk–University of Copenhagen–UC Berkeley collaboration, developing and applying targeted learning methods for causal inference, together with the accompanying software, to observational and randomized studies led by Novo Nordisk and to Danish registries with complex longitudinal data.

Beyond his dissertation, Yi co-developed Deep LTMLE, which pairs a transformer neural network with TMLE to estimate expected outcomes under treatment rules that adapt to a patient's evolving history over time, and contributed to density estimation with the highly adaptive lasso. Earlier, Yi worked with Maya Petersen on the Berkeley COVID-19 Safe Campus Initiative, contributing data analysis and dashboards and co-authoring several papers.

Yi earned his PhD in Biostatistics from UC Berkeley in 2026 and BS degrees in Mathematics (Honors) and Data Analytics, summa cum laude, from The Ohio State University in 2019.