Fast optimization via randomized numerical linear algebra
We introduce RandomizedPreconditioners.jl, a package for preconditioning linear systems using randomized numerical linear algebra. Crucially, our preconditioners do not require a priori knowledge of structure present in the linear system, making them especially useful for general-purpose algorithms. We demonstrate significant speedups of positive semidefinite linear system solves, which we use to build fast constrained optimization solvers.
