Fall 2018

Robust List Decoding of Spherical Gaussians

Monday, Oct. 29, 2018 9:30 am10:10 am PDT

Add to Calendar


Daniel Kane (UC San Diego)

We discuss new techniques for approximating the mean of a Gaussian in the presence of a large fraction of adversarial errors. We show that by taking advantage of higher moments of these distributions, we can obtain errors close to the information-theoretic optimum, and present an application of this to learning mixtures of spherical Gaussians.