arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2302.01463cs.LG
收藏

线性相关噪声下的梯度下降:理论及其在差分隐私中的应用

Gradient Descent with Linearly Correlated Noise: Theory and Applications to Differential Privacy

EPFL(瑞士洛桑联邦理工学院);Google Research(谷歌研究院)

Anastasia Koloskova, Ryan McKenna, Zachary Charles, Keith Rush, Brendan McMahan

AI总结:

本文研究差分隐私优化中线性相关噪声对梯度下降的影响,提出简化分析框架并给出更紧的理论界,据此设计新的矩阵分解方法以提升差分隐私优化性能。

摘要:

我们研究线性相关噪声下的梯度下降。我们的工作受到近期用于差分隐私(DP)优化的实用方法(如 DP-FTRL)的启发,这些方法在隐私放大技术不可行的场景(如联邦学习)中表现出强大性能。这些方法通过矩阵分解机制注入隐私噪声,使噪声在迭代过程中线性相关。我们提出一个简化设置,提炼这些方法的关键方面,并隔离线性相关噪声的影响。我们分析该设置下梯度下降在凸函数和非凸函数上的行为。我们的分析可证明比先前工作更紧,并精确恢复多个重要的特殊情形(包括反相关扰动梯度下降)。我们利用这些结果开发新的、有效的差分隐私优化矩阵分解方法,并从理论和实验上强调这些分解方法的优势。

英文摘要:

We study gradient descent under linearly correlated noise. Our work is motivated by recent practical methods for optimization with differential privacy (DP), such as DP-FTRL, which achieve strong performance in settings where privacy amplification techniques are infeasible (such as in federated learning). These methods inject privacy noise through a matrix factorization mechanism, making the noise linearly correlated over iterations. We propose a simplified setting that distills key facets of these methods and isolates the impact of linearly correlated noise. We analyze the behavior of gradient descent in this setting, for both convex and non-convex functions. Our analysis is demonstrably tighter than prior work and recovers multiple important special cases exactly (including anticorrelated perturbed gradient descent). We use our results to develop new, effective matrix factorizations for differentially private optimization, and highlight the benefits of these factorizations theoretically and empirically.

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。

↑