Differentially Private Stochastic Optimization: New Results in Convex and Non-Convex Settings

被引:0
|
作者
Bassily, Raef [1 ]
Guzman, Cristobal [2 ,3 ]
Menart, Michael [4 ]
机构
[1] Ohio State Univ, Translat Data Analyt Inst TDAI, Dept Comp Sci & Engn, Columbus, OH 43210 USA
[2] Univ Twente, Dept Appl Math, Enschede, Netherlands
[3] Pontificia Univ Catolica Chile, Inst Math & Comput Eng, Santiago, Chile
[4] Ohio State Univ, Dept Comp Sci & Engn, Columbus, OH 43210 USA
关键词
MINIMIZATION; NONSMOOTH;
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
We study differentially private stochastic optimization in convex and non-convex settings. For the convex case, we focus on the family of non-smooth generalized linear losses (GLLs). Our algorithm for the l(2) setting achieves optimal excess population risk in near-linear time, while the best known differentially private algorithms for general convex losses run in super-linear time. Our algorithm for the l(1) setting has nearly-optimal excess population risk (O) over tilde (root logd/n epsilon), and circumvents the dimension dependent lower bound of [AFKT21] for general non-smooth convex losses. In the differentially private non-convex setting, we provide several new algorithms for approximating stationary points of the population risk. For the l1-case with smooth losses and polyhedral constraint, we provide the first nearly dimension independent rate, (O) over tilde (log2/3 d/(n epsilon)1/3) in linear time. For the constrained l(2)-case with smooth losses, we obtain a linear-time algorithm with rate (O) over tilde (1/n(1/3) + d(1/5)/(n epsilon)2/5) . Finally, for the l(2)-case we provide the first method for non-smooth weakly convex stochastic optimization with rate (O) over tilde (1/n(1/4) + d(1/6) (n epsilon)(1/3)) which matches the best existing non-private algorithm when d = O(root n). We also extend all our results above for the non-convex l(2) setting to the l p setting, where 1 < p <= 2, with only polylogarithmic (in the dimension) overhead in the rates.
引用
收藏
页数:13
相关论文
共 50 条
  • [21] Convex and Non-convex Optimization Under Generalized Smoothness
    Li, Haochuan
    Qian, Jian
    Tian, Yi
    Rakhlin, Alexander
    Jadbabaie, Ali
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 36 (NEURIPS 2023), 2023,
  • [22] Riemannian Stochastic Recursive Momentum Method for non-Convex Optimization
    Han, Andi
    Gao, Junbin
    PROCEEDINGS OF THE THIRTIETH INTERNATIONAL JOINT CONFERENCE ON ARTIFICIAL INTELLIGENCE, IJCAI 2021, 2021, : 2505 - 2511
  • [23] Differentially Private Stochastic Convex Optimization under a Quantile Loss Function
    Chen, Du
    Chua, Geoffrey A.
    INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 202, 2023, 202
  • [24] Optimal, Stochastic, Non-smooth, Non-convex Optimization through Online-to-Non-convex Conversion
    Cutkosky, Ashok
    Mehta, Harsh
    Orabona, Francesco
    INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 202, 2023, 202
  • [25] A new heuristic approach for non-convex optimization problems
    Toscano, R.
    Lyonnet, P.
    INFORMATION SCIENCES, 2010, 180 (10) : 1955 - 1966
  • [26] Private (Stochastic) Non-Convex Optimization Revisited: Second-Order Stationary Points and Excess Risks
    Ganesh, Arun
    Liu, Daogao
    Oh, Sewoong
    Thakurta, Abhradeep
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 36 (NEURIPS 2023), 2023,
  • [27] Nesting of non-convex figures in non-convex contours
    Vinade, C.
    Dias, A.
    Informacion Tecnologica, 2000, 11 (01): : 149 - 156
  • [28] Momentum Aggregation for Private Non-convex ERM
    Tran, Hoang
    Cutkosky, Ashok
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 35, NEURIPS 2022, 2022,
  • [29] Stochastic Optimization for Learning Non-convex Linear Support Vector Machines
    Chen, Jifei
    Wang, Jiabao
    Zhang, Yafei
    Lu, Jianjiang
    Li, Yang
    2012 IEEE INTERNATIONAL CONFERENCE ON INTELLIGENT CONTROL, AUTOMATIC DETECTION AND HIGH-END EQUIPMENT (ICADE), 2012, : 35 - 39
  • [30] Gradient Methods for Non-convex Optimization
    Jain, Prateek
    JOURNAL OF THE INDIAN INSTITUTE OF SCIENCE, 2019, 99 (02) : 247 - 256