36-742 Statistics meets Optimization: Approximate Message Passing Algorithm

6 units, Carnegie Mellon University

In this mini, we focus our attention on the recent development of the approximate message passing algorithm. We follow a rigorous approach that builds upon ideas from statistical physics, information theory and graphical models, and is based on the analysis of an highly efficient re construction algorithm. We start with some basics for the probability graphical model, introduce the message passing algorithm and motivate the AMP algorithm along the way. Then we will discuss the exact asymptotic characterization in terms of the socalled state evolution and talk about the applications in LASSO and more generally, highdimensional robust Mestimation.

Prerequisites: none listed

Courseweave needs JavaScript to show sections, workloads and planning tools.