Privacy-Preserving Frank-Wolfe on Shuffle Model
作者机构:School of Mathematics Northwest University School of Mathematics and Information Science Baoji University of Arts and Sciences China Southern Power Grid Artificial Intelligence Technology Company Limited
出 版 物:《Acta Mathematicae Applicatae Sinica》 (应用数学学报(英文版))
年 卷 期:2024年第4期
页 面:887-907页
核心收录:
学科分类:12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 081104[工学-模式识别与智能系统] 08[工学] 0839[工学-网络空间安全] 0835[工学-软件工程] 0811[工学-控制科学与工程] 081201[工学-计算机系统结构] 0812[工学-计算机科学与技术(可授工学、理学学位)]
基 金:supported by the National Natural Science Foundation of China (No. U1811461 12326615)
摘 要:In this paper,we design the differentially private variants of the classical Frank-Wolfe algorithm with shuffle model in the optimization of machine *** weak assumptions and the generalized linear loss (GLL) structure,we propose a noisy Frank-Wolfe with shuffle model algorithm (NoisyFWS) and a noisy variance-reduced Frank-Wolfe with the shuffle model algorithm (NoisyVRFWS) by adding calibrated laplace noise under shuffling scheme in the?p(p∈[1,2])-case,and study their privacy as well as utility guarantees for the H?lder smoothness *** particular,the privacy guarantees are mainly achieved by using advanced composition and privacy amplification by *** utility bounds of the Noisy FWS and NoisyVRFWS are analyzed and obtained the optimal excess population risks ■ and ■with gradient complexity ■ for ■.It turns out that the risk rates under shuffling scheme are a nearly-dimension independent rate,which is consistent with the previous work in some *** addition,there is a vital tradeoff between (α,L)-H?lder smoothness GLL and the gradient *** linear gradient complexity O(n) is showed by the parameter α=1.