arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2311.00531math.OC
收藏

高斯平滑提升随机梯度的性能

Improved Performance of Stochastic Gradients with Gaussian Smoothing

Andrew Starnes, Clayton Webster

AI总结:

本文提出将高斯平滑应用于深度学习中的SGD和Adam优化算法,通过衰减微小波动降低陷入较差局部极小值的风险,并推导网络的高斯平滑损失函数以提升计算效率,实验证明其优于未平滑版本。

摘要:

本文形式化并分析了高斯平滑在深度学习中两种重要优化方法上的应用:随机梯度下降(GSmoothSGD)和Adam(GSmoothAdam)。通过衰减微小波动,高斯平滑降低了基于梯度的算法收敛到较差局部极小值的风险。这些方法在提升对噪声鲁棒性的同时简化了损失景观,并改善了泛化能力,帮助基础算法更有效地收敛到全局极小值。现有方法通常依赖零阶近似,由于自动微分效率低下而增加了训练时间。为此,我们推导了前馈和卷积网络的高斯平滑损失函数,提升了计算效率。数值实验表明,我们的平滑算法相较于未平滑的版本具有更优的性能,证实了其理论优势。

英文摘要:

This paper formalizes and analyzes Gaussian smoothing applied to two prominent optimization methods: Stochastic Gradient Descent (GSmoothSGD) and Adam (GSmoothAdam) in deep learning. By attenuating small fluctuations, Gaussian smoothing lowers the risk of gradient-based algorithms converging to poor local minima. These methods simplify the loss landscape while boosting robustness to noise and improving generalization, helping base algorithms converge more effectively to global minima. Existing approaches often rely on zero-order approximations, which increase training time due to inefficiencies in automatic differentiation. To address this, we derive Gaussian-smoothed loss functions for feedforward and convolutional networks, improving computational efficiency. Numerical experiments demonstrate the enhanced performance of our smoothing algorithms over unsmoothed counterparts, confirming the theoretical benefits.

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。

↑