Fast and Simple PCA via Convex Optimization

The problem of principle component analysis (PCA) is traditionally solved by spectral or algebraic methods. We show how computing the leading principal component could be reduced to solving a textitsmall number of well-conditioned it convex optimization problems. This gives rise to a new efficient method for PCA based on recent advances in stochastic methods for convex optimization. In particular we show that given a d× d matrix X = frac1nsumi=1nxixitop with top eigenvector u and top eigenvalue λ1 it is possible to: beginitemize item compute a unit vector w such that (wtopu)2 ≥ 1-ε in tildeOleft(fracdδ2+Nright) time, where δ = λ1 - λ2 and N is the total number of non-zero entries in x1,...,xn, item compute a unit vector w such that wtopXw ≥ λ1-ε in tildeO(d/ε2) time. enditemize To the best of our knowledge, these bounds are the fastest to date for a wide regime of parameters. These results could be further accelerated when δ (in the first case) and ε (in the second case) are smaller than √d/N.