arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2006.07037math.OCcs.LG
收藏

自适应梯度法在有限时期后可证明比SGD快

Adaptive Gradient Methods Can Be Provably Faster than SGD after Finite Epochs

Bytedance AI Lab(字节跳动人工智能实验室);Peking University(北京大学);Ohio State University(俄亥俄州立大学)

Xunpeng Huang, Hao Zhou, Runxin Xu, Zhe Wang, Lei Li

摘要:

自适应梯度法因其高效性而受到机器学习界的广泛关注。然而,它们在实践中的加速效果,特别是在神经网络训练中,很难从理论上分析。理论收敛结果和实际性能之间的巨大差距阻碍了对现有优化器的进一步理解和更先进的最优化方法的发展。在本文中,我们提供了自适应梯度法一个新的分析与额外的温和的假设,并修改AdaGrad为\radagrad匹配一个更好的可证明的收敛速度。为了找到一个$ε$ -近似的一阶驻点在非凸目标,我们证明随机洗牌\radagrad实现$\tilde{O}(T^{-1/2})$收敛速度,这是显着改善的因素$\tilde{O}(T^{-1/4})$和$\tilde{O}(T^{-1/6})$相比,现有的自适应梯度法和随机洗牌SGD,分别。据我们所知,这是第一次证明,自适应梯度法可以确定性地比SGD有限时期后更快。此外,我们进行了全面的实验,以验证额外的温和的假设和加速效果受益于二阶矩和随机洗牌。

英文摘要:

Adaptive gradient methods have attracted much attention of machine learning communities due to the high efficiency. However their acceleration effect in practice, especially in neural network training, is hard to analyze, theoretically. The huge gap between theoretical convergence results and practical performances prevents further understanding of existing optimizers and the development of more advanced optimization methods. In this paper, we provide adaptive gradient methods a novel analysis with an additional mild assumption, and revise AdaGrad to \radagrad for matching a better provable convergence rate. To find an $ε$-approximate first-order stationary point in non-convex objectives, we prove random shuffling \radagrad achieves a $\tilde{O}(T^{-1/2})$ convergence rate, which is significantly improved by factors $\tilde{O}(T^{-1/4})$ and $\tilde{O}(T^{-1/6})$ compared with existing adaptive gradient methods and random shuffling SGD, respectively. To the best of our knowledge, it is the first time to demonstrate that adaptive gradient methods can deterministically be faster than SGD after finite epochs. Furthermore, we conduct comprehensive experiments to validate the additional mild assumption and the acceleration effect benefited from second moments and random shuffling.

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。

↑