Dr. Dror Baron has been awarded $422,732 by the National Science Foundation for research on CIF: Small: Universal Signal Estimation from Noisy Measurements.

The award will run from September 1st, 2012 to August 31st, 2015.

Research Abstract

Motivation: A ubiquitous feature in many signal processing systems is to learn the input statistics from historical data. In these systems, Bayesian methods perform statistically optimal signal processing. However, there are applications including file compression, speech recognition, network monitoring, and compressed sensing in which it might be impractical to learn the statistics a priori. In such applications, a statistical approach that adapts to the data at hand must be used.
The information theory community has championed the use of universal algorithms, they achieve the best possible statistical performance asymptotically despite not knowing the input statistics.
These algorithms have had tremendous impact in lossless compression, where the goal is to describe data as succinctly as possible while allowing a decoder to reproduce the input perfectly. In sharp contrast, universal algorithms have had little impact on other areas.