Abstract

The sum-of-squares hierarchy has led to new polynomial time algorithms for many problems, but solving semidefinite programs is often too computationally intensive to be practical. In this talk I'll describe a couple of works in which we bypass the use of constant-round SOS SDPs, and instead give lightweight (and sometimes near-linear time) spectral algorithms based on the SOS analyses.

Based on joint works with Sam Hopkins, Jonathan Shi and David Steurer.

Video Recording