arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2402.03210math.OC
收藏

随机凸优化的通用梯度方法

Universal Gradient Methods for Stochastic Convex Optimization

Anton Rodomanov, Ali Kavis, Yongtao Wu, Kimon Antonakopoulos, Volkan Cevher

AI总结:

本研究针对随机凸优化问题,提出基于梯度差与迭代点差组合累积调整步长的通用梯度方法,可自动适配噪声与赫尔德光滑性,达当前最优最坏情况收敛率,还给出了具最优效率估计的通用快速梯度方法。

摘要:

我们提出了适用于随机凸优化(Stochastic Convex Optimization,SCO)的通用梯度方法。我们的算法无需预先知晓具体设置,不仅能自动适配预言机的噪声,还能自动适配目标函数的赫尔德(Hölder)光滑性。方法的核心是一种在随机梯度法(Stochastic Gradient Method,SGD)中调整步长系数的全新策略。与累积梯度范数的AdaGrad不同,我们的通用梯度方法累积的是梯度差与迭代点差的恰当组合。所得算法对整个赫尔德函数类都具备当前最优的最坏情况收敛速率保证,尤其涵盖了非光滑函数和具有利普希茨(Lipschitz)连续梯度的函数。我们还提出了适用于SCO的通用快速梯度方法,它具有最优的效率估计。

英文摘要:

We develop universal gradient methods for Stochastic Convex Optimization (SCO). Our algorithms automatically adapt not only to the oracle's noise but also to the Hölder smoothness of the objective function without a priori knowledge of the particular setting. The key ingredient is a novel strategy for adjusting step-size coefficients in the Stochastic Gradient Method (SGD). Unlike AdaGrad, which accumulates gradient norms, our Universal Gradient Method accumulates appropriate combinations of gradient- and iterate differences. The resulting algorithm has state-of-the-art worst-case convergence rate guarantees for the entire Hölder class including, in particular, both nonsmooth functions and those with Lipschitz continuous gradient. We also present the Universal Fast Gradient Method for SCO enjoying optimal efficiency estimates.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑