Spectral Methods and the Landscapes of High-Dimensional Inference Problems

Yue Lu
Harvard University

Spectral methods are widely used in nonconvex optimization approaches to signal estimation. Examples include phase retrieval, blind deconvolution, and low-rank matrix/tensor recovery. In this talk, I will present our work on precise asymptotic characterizations of the performance of the spectral methods and on their connections to the landscapes of high-dimensional inference problems.

Back to Workshop IV: Using Physical Insights for Machine Learning