We study $k$-GenEV, the problem of finding the top $k$ generalized eigenvectors, and $k$-CCA, the problem of finding the top $k$ vectors in canonical-correlation analysis. We propose algorithms $\mathtt{ LazyEV}$ and $\mathtt{ LazyCCA}$ to solve the two problems with running times linearly dependent on the input size and on $k$. 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 $k$-GenEV or $k$-CCA. We also provide the first gap-free results, which provide running times that depend on $1/\sqrt{ \varepsilon}$ rather than the eigengap.
Categories: math.OC cs.DS cs.LG stat.ML