Fast Stochastic Bregman Gradient Methods: Sharp Analysis and Variance Reduction
From MaRDI portal
Abstract: We study the problem of minimizing a relatively-smooth convex function using stochastic Bregman gradient methods. We first prove the convergence of Bregman Stochastic Gradient Descent (BSGD) to a region that depends on the noise (magnitude of the gradients) at the optimum. In particular, BSGD with a constant step-size converges to the exact minimizer when this noise is zero (emph{interpolation} setting, in which the data is fit perfectly). Otherwise, when the objective has a finite sum structure, we show that variance reduction can be used to counter the effect of noise. In particular, fast convergence to the exact minimizer can be obtained under additional regularity assumptions on the Bregman reference function. We illustrate the effectiveness of our approach on two key applications of relative smoothness: tomographic reconstruction with Poisson noise and statistical preconditioning for distributed optimization.
This page was built for publication: Fast Stochastic Bregman Gradient Methods: Sharp Analysis and Variance Reduction
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6365707)
