Title: Sequential Testing in High Dimensions
Sequential methods make use of information as it becomes available, creating an interactive connection between a sampling procedure and information gathered by that procedure. In this talk we explore sequential methods applied to sparse recovery problems, motivated by applications in both communications and biology. Surprisingly, sequential methods can result in a large reduction in the sample size needed to recover a sparse signal. Discovery Building, Orchard View Room Krishna Sridhar, Matt Malloy
More specifically, we consider an n-dimensional vector x whose elements are drawn from one of two distributions, p0 and p1. Most of the elements are drawn from p0, but a small number, s, are drawn from p1. The goal is to identify the unknown locations of this sparse subset. Non-sequential testing schemes require at least log(n)/D(p1||p0) samples per dimension, where D(p0||p1) is the KL-divergence from p1 to p0. In the high-dimensional and sparse regimes (n large, s<