Title: Copula Applications and Reinforcement Learning for Personalized Treatment Policies in the Context of High-Dimensional Missing and Censored Data

Presented by Dr. Jong-Min Kim
Professor of Statistics
University of Minnesota, Morris

Abstract: Copula is a valuable statistical method for elucidating the dependence structure between variables by mitigating the influence of marginals. This talk aims to delve into its significance. The main points of this talk are as follows: Firstly, we introduce a copula method that captures directional dependence, revealing direct interactions among fMRI data. Secondly, we apply the copula dynamic correlation coefficient method, functional principal component analysis, and graphical visualization to US Covid-19 mortality data to demonstrate the practical applications of the copula approach. Thirdly, the talk introduces an innovative algorithm that integrates Q-Learning with imputed survival rewards, addressing challenges arising from incomplete patient information and censored event times. The algorithm initializes Q-values, defines state and action spaces, and employs Boruta variable selection and multivariate imputation via chained equations for handling missing data. Leveraging the Buckley-James method, it estimates imputed survival times and updates Q-values accordingly. The final phase involves extracting personalized treatment policies that maximize the expected total imputed survival reward.

A seminar tea will be held at 9:30 a.m. in University Office Plaza, Room 240. All are Welcome.