Statistics Seminar: Yuhong Yang
Speaker: Yuhong Yang
Title: Personalized Treatment Allocations Based on Multi-Armed Bandits with Covariates
Abstract: In practice of medicine (and many other fields), multiple treatments (in a broad sense) are often available to treat individual patients (or subjects, customers etc). The task of online identification of the best treatment for a specific patient is very challenging due to patient inhomogeneity. Multi-armed bandits with covariates (MABC), also called contextual bandits, provide a framework for designing effective treatment allocation rules in a way that integrates the learning from experimentation with maximizing the benefits to the patients along the process.
In this talk, we review basics of MABC and present some randomized (or epsilon-greedy) non-parametric strategies to achieve strongly consistent or minimax optimal treatment allocations, possibly with delayed observations of the outcomes. Simulations and a real data example are given to demonstrate the performance of the proposed MABC methods.
The talk is based on joint work with Dan Zhu, Wei Qian and Sakshi Arya.
A virtual social will take place from 11:30-12:00, following this seminar. To join, visit z.umn.edu/STATSeminarSocial
NOTE: You will need to use a Chrome or Firefox browser to access this virtual platform. The link may not open if you sign in using a mobile device or tablet. The virtual space will not open until approximately 11:30.