|
Activity Number:
|
445
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 2, 2016 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistical Learning and Data Science
|
|
Abstract #319779
|
|
|
Title:
|
Regret Bounds for a Thompson Sampling Algorithm with Application to Emerging Infectious Disease
|
|
Author(s):
|
Tao Hu* and Eric Laber
|
|
Companies:
|
North Carolina State University and North Carolina State University
|
|
Keywords:
|
data-driven management ;
Thompson sampling ;
convergence rates ;
online estimation ;
multi-armed bandit ;
emerging infectious disease
|
|
Abstract:
|
Emerging infectious diseases are a persistent threat to public health and economies across the world. Managing an emerging infectious disease is challenging as little or no information about disease dynamics or intervention effectiveness is known at outbreak. Thus, data-driven management of an emerging infectious disease must be done online wherein estimation of disease dynamics and treatment allocation must be done simultaneously and in real-time. There is a trade-off between treatment allocations that lead to a large gain in information about disease dynamics with treatment allocations that would minimize spread based on current estimates of disease dynamics. We balance this trade-off by recasting treatment allocation decisions as a multi-armed bandit problem and applying a variant of Thompson sampling. We prove that Thompson sampling is consistent for an optimal treatment allocation strategy in this setting and provide rates of convergence. This methodology is illustrated using a model for the spread of Cholera in Zimbabwe.
|
Authors who are presenting talks have a * after their name.
