22
50

Doubly Accelerated Methods for Faster CCA and Generalized Eigendecomposition

Abstract

We study kk-GenEV, the problem of finding the top kk generalized eigenvectors, and kk-CCA, the problem of finding the top kk vectors in canonical-correlation analysis. We propose algorithms LazyEV\mathtt{LazyEV} and LazyCCA\mathtt{LazyCCA} to solve the two problems with running times linearly dependent on the input size and on kk. Furthermore, our algorithms are DOUBLY-ACCELERATED: our running times depend only on the square root of the matrix condition number, and on the square root of the eigengap. This is the first such result for both kk-GenEV or kk-CCA. We also provide the first gap-free results, which provide running times that depend on 1/ε1/\sqrt{\varepsilon} rather than the eigengap.

View on arXiv
Comments on this paper