Fast Stochastic Bregman Gradient Methods: Sharp Analysis and Variance Reduction
arXiv:2104.09813
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.
Cited by in corpus (4)
- Local convexity of the TAP free energy and AMP convergence for Z2-synchronization
- A Bregman Learning Framework for Sparse Neural Networks
- Concentration of Non-Isotropic Random Tensors with Applications to Learning and Empirical Risk Minimization
- Deep Bregman Divergence for Contrastive Learning of Visual Representations