Mathematics Colloquium
Speaker: Parthe Pandit, IIT Bombay
Host: Parthanil Roy
Title: Precise Asymptotics of Statistical Linear Inverse Problems in High Dimensions
Time, day and date: 4:15:00 PM - 5:15:00 PM, Wednesday, October 14
Venue: Ramanujan Hall
Abstract: This is a tutorial-style talk on a class of algorithms studied under the name of Approximate Message Passing. These algorithms have a scalar State Evolution which enables characterizing the asymptotic behavior of their fixed points in a precise manner in the proportional high dimensional regime. This enables rigorous analysis of a large class of optimization formulations such as regularized M-estimation for generalized linear models, and structural risk minimization in machine learning.