YSPH Biostatistics Seminar: "Nonparametric Finite-Horizon Reinforcement Learning For Right-Censored Outcomes"

NOTE: BIS 526 students are required to attend in person (47 College St., Room 106A). All others are requested to attend via Zoom.

Speaker- Michael R. Kosorok, Ph.D.

Title- "Nonparametric finite-horizon reinforcement learning for right-censored outcomes"

Abstract

In the presentation, we describe a new, nonparametric off-policy, finite-horizon reinforcement learning methodology for right censored data. The method is flexible enough to allow for censoring which is conditionally independent given the available history as well as for visit times that may be intrinsically dependent on treatment response. We show that the proposed method is asymptotically consistent and has some demonstrable advantages over alternative approaches. We evaluate the method in both simulations and in the analysis of data from a hybrid randomized and observational study of leukemia treatment.