Sequential Decision Making with Uncertainty: Contextual Bandits as a Missing Data Problem

Fri, Aug 18, 2023, 09:00 – 10:00AM (PDT), 12:00 – 1:00PM (EST)

Abstract: 

The contextual bandit is a sequential decision-making problem where a learner repeatedly selects an arm based on contextual information and receives a reward as partial feedback associated with the chosen arm only. Bandit methodologies have been appliedin enhancing health outcomes through self-guided digital interventions. The learner's objective is to maximize cumulative rewards over time while facing uncertainty regarding the underlying true reward-generating mechanism. We cast the contextual bandit as a missing data problem and introduce novel algorithms. We discuss their advantages from a non-asymptotic perspective. Through empirical analysis, we demonstrate the superior performance of our proposed algorithm compared to existing methods.

Speaker:

Myunghee Cho Paik has been Professor of Statistics at Seoul National University since 2012. She received a Ph.D. in Biostatistics from University of Pittsburgh in 1987. She served as a professor in the Department of Biostatistics at Columbia University from 1988 to 2012 before joining the Department of Statistics, Seoul National University. Her research area includes longitudinal data, missing data, and sequential decision-making.

Strategic Alliance: