CAREER: Toward a Unified Approach to Universality in Information Processing

Stanford University, Stanford CA

Investigators

Abstract

A major challenge in signal processing, communication, and compression applications is the construction of schemes that approach the theoretical performance limits. The fact that a priori knowledge of source and channel characteristics is rarely available in practice necessitates that these schemes be universal. While it is clear that existing universal schemes for different problems share many common ingredients, thus far they have been considered mostly separately. This project develops a unified framework encompassing theoretical, algorithmic, and implementation aspects of the universality problem. We address questions concerning universal schemes in problem areas ranging from denoising, Wyner-Ziv coding, joint source-channel coding, scanning and prediction, to reinforcement learning. We develop a unified framework for addressing such questions, which facilitates the construction of universal schemes for new problems, and leads to new insights on existing ones. On the more applied side, we are developing new algorithmic paradigms that have attractive complexity, and applying them in practice. We are also developing analytic tools for assessing the performance of the new algorithms.