Online Program Home
My Program

Abstract Details

Activity Number: 213 - Sequential Decision Making and Causal Inference
Type: Invited
Date/Time: Monday, July 29, 2019 : 2:00 PM to 3:50 PM
Sponsor: IMS
Abstract #300465
Title: Learning to Personalize from Observational Data Under Unobserved Confounding
Author(s): Nathan Kallus*
Companies: Cornell University and Cornell Tech
Keywords: personalization; observational data; policy learning; machine learning; confounding; sensitivity analysis
Abstract:

Since observational data no matter how rich will have some unobserved confounding, methods for individualized decision making that assume unconfoundedness may unknowingly lead to harm and are therefore often unusable in practice. We instead propose a new minimax-regret formulation based on marginal sensitivity analysis, sharp policy evaluation, and robust optimization; prove that the policy we get is (almost) always guaranteed to improve on existing standards of care; and develop specialized optimization algorithms to find this policy. A personalized medicine application in stroke treatment demonstrates the power of our approach and danger of previous ones.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program