无惧选择：几乎所有的小批量调度都可以最优泛化（英文版）

英文标题：Select without Fear: Almost All Mini-Batch Schedules Generalize Optimally中文摘要：本研究建立了优化算法，分析了批处理的优点，证明了基于批处理训练的渐进误差上下界。英文摘要：We establish matching upper and lower generalization error bounds formini-batch Gradient Descent (GD) training with either deterministic orstochastic, data-independent, but otherwise arbitrary batch selection rules. Weconsider smooth Lipschitz-convex/nonconvex/strongly-convex loss functions, andshow that classical upper bounds for Stochastic GD (SGD) also hold verbatim