11
57

Sublinear Time Low-Rank Approximation of Positive Semidefinite Matrices

Abstract

We show how to compute a relative-error low-rank approximation to any positive semidefinite (PSD) matrix in sublinear time, i.e., for any n×nn \times n PSD matrix AA, in O~(npoly(k/ϵ))\tilde O(n \cdot poly(k/\epsilon)) time we output a rank-kk matrix BB, in factored form, for which ABF2(1+ϵ)AAkF2\|A-B\|_F^2 \leq (1+\epsilon)\|A-A_k\|_F^2, where AkA_k is the best rank-kk approximation to AA. When kk and 1/ϵ1/\epsilon are not too large compared to the sparsity of AA, our algorithm does not need to read all entries of the matrix. Hence, we significantly improve upon previous nnz(A)nnz(A) time algorithms based on oblivious subspace embeddings, and bypass an nnz(A)nnz(A) time lower bound for general matrices (where nnz(A)nnz(A) denotes the number of non-zero entries in the matrix). We prove time lower bounds for low-rank approximation of PSD matrices, showing that our algorithm is close to optimal. Finally, we extend our techniques to give sublinear time algorithms for low-rank approximation of AA in the (often stronger) spectral norm metric AB22\|A-B\|_2^2 and for ridge regression on PSD matrices.

View on arXiv
Comments on this paper