UW News

November 3, 2016

Electrical engineering lecture series to explore compressed sensing

David Donoho

David Donoho

In the last decade, the signal-processing technique of compressed sensing has delivered notable speedups in medical imaging, from pediatric MRI (magnetic resonance imaging) to dynamic cardiac imaging. It also advanced scientific signal processing, including NMR (nuclear magnetic resonance) spectroscopy and compressed genotyping.

The compressed sensing terminology arose in applied mathematics and information theory, where theory suggested that massive speedups ought to be possible, provided one could make randomly-selected measurements. Practitioners have made striking speedups, yet often without using the randomly-selected measurements that theorists were insisting on.

As part of the 2016 Lytle Lecture Series hosted by the University of Washington Department of Electrical Engineering, Stanford University Anne T. and Robert M. Bass Professor of Humanities and Sciences and Professor of Statistics David Donoho will discuss how theorists and practitioners can bridge this gap.

“Theorists have favored using random projection matrices to make measurements. However, practitioners have always objected that random projections are not available, because the technology does not allow them,” Donoho said, pointing specifically to gaps in MR imaging and in genotyping.  “And yet they succeed. Theorists need to understand what is technologically possible, and develop theory that exploits it.”

Donoho has made fundamental contributions to theoretical and computational statistics, as well as to signal processing and harmonic analysis. His algorithms have contributed significantly to our understanding of the maximum entropy principle, the structure of robust procedures and sparse data description. Over the last 30 years of research, Donoho has focused on the overall “big challenge” — to recover, or simplify, information from symbols and images.

“We’re in a media-saturated era, where images and signals are part of everything we do,” Donoho said.  “Just look at the amount of data transfer in Google and Facebook. The big challenge of our era is to tease meaningful signals out of raw data.  As a theorist, I seek to exploit the simple structure of many important signals and images to deliver better results.”

Donoho notes that signal processing is everywhere, from medical imaging to cellphone reception. His Lytle Lecture will survey recent developments that are bringing this theory and practice closer together.

Donoho will give a free public talk entitled “Compressed Sensing: From Theory to Practice” on Monday, Nov. 7, from 3:30 to 4:30 p.m. in the Paul Allen Center Atrium.

On Tuesday, Nov. 8, Donoho will give a second talk to the research community. The talk, “High-Dimensional Statistics in Light of the Spiked Covariance Model,” will be from 10:30 to 11:30 a.m. in EEB 105 in the Paul G. Allen Center. This lecture is also free and open to the public.

The lectures will also be livestreamed and will be available for later viewing on the department’s YouTube channel.

Among many awards, Donoho was named a MacArthur Fellow, was elected as a SIAM (Society for Industrial and Applied Mathematics) Fellow, received the Shaw Prize for Mathematics and won the COPSS Presidents’ Award. He also received the Norbert Wiener Prize in Applied Mathematics and John von Neumann Prize.

The Lytle Lecture honors the late Professor Dean W. Lytle, who began his career as an assistant professor in 1958 in the UW Department of Electrical Engineering. Professor Lytle’s teaching, research and high-impact consulting reached from communications, networks and probability to signal processing.

For more information, contact Annie Pellicciotti at apell@uw.edu.