Without-Replacement Sampling for Stochastic Gradient Methods

摘要

Stochastic gradient methods for machine learning and optimization problems are usually analyzed assuming data points are sampled