arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2408.07503math.OCcs.LG
收藏

通过异步小批量实现任意延迟下的更快随机优化

Faster Stochastic Optimization with Arbitrary Delays via Asynchronous Mini-Batching

Blavatnik School of Computer Science, Tel Aviv University(塔夫茨大学布劳维特克计算机科学学院);Tel Aviv University(塔夫茨大学);Google Research Tel Aviv(谷歌研究团队(特拉维夫))

Amit Attia, Ofir Gaash, Tomer Koren

AI总结:

针对异步随机优化中任意延迟问题,提出基于异步小批量与梯度过滤的通用转换方法,实现依赖分位数延迟的收敛速率,并自适应所有分位数,显著优于现有平均延迟结果。

摘要:

我们考虑异步随机优化问题,其中优化算法基于目标的过时随机梯度进行更新,这些梯度受到任意(可能是对抗性的)延迟序列的影响。我们提出一种程序,对于任意给定的 $q \in (0,1]$,将任何标准随机一阶方法转换为异步方法,其收敛保证取决于序列的 $q$-分位数延迟。该方法在非凸问题中实现形如 $O(τ_q/qT+σ/\sqrt{qT})$ 的收敛速率,在凸光滑问题中实现形如 $O(τ_q^2/(q T)^2+σ/\sqrt{qT})$ 的收敛速率,其中 $τ_q$ 是 $q$-分位数延迟,推广并改进了依赖于平均延迟的现有结果。我们进一步展示一种方法,无需任何关于延迟的先验知识,即可同时自动适应所有分位数,在非凸问题中实现形如 $O(\inf_{q} τ_q/qT+σ/\sqrt{qT})$ 的收敛速率,在凸光滑问题中实现形如 $O(\inf_{q} τ_q^2/(q T)^2+σ/\sqrt{qT})$ 的收敛速率。我们的技术基于异步小批量,并仔细选择批量大小和过滤过时梯度。

英文摘要:

We consider the problem of asynchronous stochastic optimization, where an optimization algorithm makes updates based on stale stochastic gradients of the objective that are subject to an arbitrary (possibly adversarial) sequence of delays. We present a procedure which, for any given $q \in (0,1]$, transforms any standard stochastic first-order method to an asynchronous method with convergence guarantee depending on the $q$-quantile delay of the sequence. This approach leads to convergence rates of the form $O(τ_q/qT+σ/\sqrt{qT})$ for non-convex and $O(τ_q^2/(q T)^2+σ/\sqrt{qT})$ for convex smooth problems, where $τ_q$ is the $q$-quantile delay, generalizing and improving on existing results that depend on the average delay. We further show a method that automatically adapts to all quantiles simultaneously, without any prior knowledge of the delays, achieving convergence rates of the form $O(\inf_{q} τ_q/qT+σ/\sqrt{qT})$ for non-convex and $O(\inf_{q} τ_q^2/(q T)^2+σ/\sqrt{qT})$ for convex smooth problems. Our technique is based on asynchronous mini-batching with a careful batch-size selection and filtering of stale gradients.

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。

↑