BAYESIAN VARIABLE SELECTION STRATEGIES FOR HIGH-DIMENSIONAL DATA
MAHLET TADESSE – GEORGETOWN UNIVERSITY
The problem of variable selection in high-dimensional data has received a lot of attention over the past three decades. In this talk, I will give an overview of the two main lines of work I have been pursuing in this area. The first consists of uncovering cluster structures and identifying component-specific features by combining ideas of mixture models and variable selection. The second focuses on specifying prior distributions that integrate external knowledge or that account for the dependence structure in the data, using either spike-and-slab or continuous shrinkage priors. These strategies can increase the power to detect subtle effects without increasing the probability of false discoveries, and can improve the predictive performance.