Efficient coordinate-wise leading eigenvector computation

[edit]

Jialei Wang, Weiran Wang, Dan Garber, Nathan Srebro ;
Proceedings of Algorithmic Learning Theory, PMLR 83:806-820, 2018.

Abstract

We develop and analyze efficient "coordinate-wise" methods for finding the leading eigenvector, where each step involves only a vector-vector product. We establish global convergence with overall runtime guarantees that are at least as good as Lanczos’s method and dominate it for slowly decaying spectrum. Our methods are based on combining a shift-and-invert approach with coordinate-wise algorithms for linear regression.

Related Material