Data Processing Theorems and Applications
Speaker: Sudeep Kamath, University of California-San Diego
Department: Electrical Engineering
Location: Engineering Quadrangle B205
Date/Time: Tuesday, March 25, 2014, 4:30 p.m. - 5:30 p.m.
Understanding the amount of information loss in a randomized map is an important problem in a variety of contexts, such as stochastic simulation, learning, and computation in noisy circuits. Data processing theorems attempt to characterize this loss of information.
In this talk, we will consider so-called strong data processing inequalities. Simply stated, a data processing inequality guarantees that processing of data may only reduce its information content while a strong data processing inequality makes this precise by quantifying the loss. We will consider the question of the tightest possible strong data processing inequalities for mutual information and discuss some implications for results in the literature on multi-terminal information theory.