Abstract:

Randomized encoding is used in communication, document retrieval, dimensional reduction and even numerical analysis, with remarkable counterintuitive efficiency. Our goal is to try to explain this success, and illustrate signal processing capabilities in the encoded compressed sensing domain. In particular we indicate how to search for features in encoded samples (matched filtering) as well as methodologies to search for best basis and related basis searches. To conclude we indicate that nonlinear data modeling is feasible in this setting.