Prof. Alex Dimakis Delivered Plenary Talk at Conference on Information Sciences and Systems

Monday, March 26, 2018

Prof. Alex Dimakis of Texas ECE was one of three distinguished speakers o give a plenary talk at the 52nd annual Conference on Information Sciences and Systems (CISS) at Princeton University on March 23, 2018. Prof. Dimakis delivered a talk entitled "Generative Adversarial Networks (GANs) and Compressed Sensing." Prof. Pramod Viswanath of the University of Illinois at Urbana-Champaign and Prof. Bin Yu of the University of California, Berkeley also gave plenary talks.

CISS was hosted by the Princeon University Department of Electrical Engineering and the IEEE Information Society and featured papers and discussions "describing theoretical advances, applications, and ideas in the fields of: Information Theory, Coding Theory, Image Processing, Communications,Signal Processing, Machine Learning, Statistical Inference, Security and Privacy, Energy Systems, Networking, Systems and Control, and Biological Systems."

The abstract for Prof. Dimakis's talk is below:
The goal of compressed sensing is to estimate a vector from an underdetermined system of noisy linear measurements, by making use of prior knowledge on the structure of vectors in the relevant domain. For almost all results in this literature, the structure is represented by sparsity in a well-chosen basis. We show how to achieve guarantees similar to standard compressed sensing but without employing sparsity at all. Instead, we suppose that vectors lie near the range of a generative model, e.g. a GAN or a VAE. We show how the problems of image completion and super-resolution are special cases of our general framework. We show how to generalize the RIP condition for generative models and that random gaussian measurement matrices have this property with high probability. This research has applications to medical imaging, solving inverse problems and security in machine learning.  

(based on joint work with Ashish Bora, Ajil Jalal and Eric Price)

Funding for the UT Austin team's work was provided by NSF, an ARO Young Investigator award and an NVIDIA GPU grant.