15
128

LazySVD: Even Faster SVD Decomposition Yet Without Agonizing Pain

Abstract

We study kk-SVD that is to obtain the first kk singular vectors of a matrix AA. Recently, a few breakthroughs have been discovered on kk-SVD: Musco and Musco [1] proved the first gap-free convergence result using the block Krylov method, Shamir [2] discovered the first variance-reduction stochastic method, and Bhojanapalli et al. [3] provided the fastest O(nnz(A)+poly(1/ε))O(\mathsf{nnz}(A) + \mathsf{poly}(1/\varepsilon))-time algorithm using alternating minimization. In this paper, we put forward a new and simple LazySVD framework to improve the above breakthroughs. This framework leads to a faster gap-free method outperforming [1], and the first accelerated and stochastic method outperforming [2]. In the O(nnz(A)+poly(1/ε))O(\mathsf{nnz}(A) + \mathsf{poly}(1/\varepsilon)) running-time regime, LazySVD outperforms [3] in certain parameter regimes without even using alternating minimization.

View on arXiv
Comments on this paper