ANALYSIS OF GENERALIZED BREGMAN SURROGATE ALGORITHMS FOR NONSMOOTH NONCONVEX STATISTICAL LEARNING

被引:2
作者
She, Yiyuan [1 ]
Wang, Zhifeng [1 ]
Jin, Jiuwu [1 ]
机构
[1] Florida State Univ, Dept Stat, Tallahassee, FL 32306 USA
基金
美国国家科学基金会;
关键词
Nonconvex optimization; nonsmooth optimization; MM algorithms; Bregman divergence; statistical algorithmic analysis; momentum-based acceleration; NONCONCAVE PENALIZED LIKELIHOOD; VARIABLE SELECTION; ORACLE INEQUALITIES; CONVERGENCE; REGRESSION; SHRINKAGE; MODELS;
D O I
10.1214/21-AOS2090
中图分类号
O21 [概率论与数理统计]; C8 [统计学];
学科分类号
020208 ; 070103 ; 0714 ;
摘要
Modern statistical applications often involve minimizing an objective function that may be nonsmooth and/or nonconvex. This paper focuses on a broad Bregman-surrogate algorithm framework including the local linear approximation, mirror descent, iterative thresholding, DC programming and many others as particular instances. The recharacterization via generalized Bregman functions enables us to construct suitable error measures and establish global convergence rates for nonconvex and nonsmooth objectives in possibly high dimensions. For sparse learning problems with a composite objective, under some regularity conditions, the obtained estimators as the surrogate's fixed points, though not necessarily local minimizers, enjoy provable statistical guarantees, and the sequence of iterates can be shown to approach the statistical truth within the desired accuracy geometrically fast. The paper also studies how to design adaptive momentum based accelerations without assuming convexity or smoothness by carefully controlling stepsize and relaxation parameters.
引用
收藏
页码:3434 / 3459
页数:26
相关论文
共 50 条
[21]   Generalized Nonconvex Nonsmooth Low-Rank Matrix Recovery Framework With Feasible Algorithm Designs and Convergence Analysis [J].
Zhang, Hengmin ;
Qian, Feng ;
Shi, Peng ;
Du, Wenli ;
Tang, Yang ;
Qian, Jianjun ;
Gong, Chen ;
Yang, Jian .
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS, 2023, 34 (09) :5342-5353
[22]   Distributed Multi-block Partially Symmetric Bregman ADMM for Nonconvex and Nonsmooth Sharing Problem [J].
Cui, Tian-Tian ;
Dang, Ya-Zheng ;
Gao, Yan .
JOURNAL OF THE OPERATIONS RESEARCH SOCIETY OF CHINA, 2025,
[23]   Accelerated gradient methods for sparse statistical learning with nonconvex penalties [J].
Yang, Kai ;
Asgharian, Masoud ;
Bhatnagar, Sahir .
STATISTICS AND COMPUTING, 2024, 34 (01)
[24]   Inexact Block Coordinate Descent Algorithms for Nonsmooth Nonconvex Optimization [J].
Yang, Yang ;
Pesavento, Marius ;
Luo, Zhi-Quan ;
Ottersten, Bjorn .
IEEE TRANSACTIONS ON SIGNAL PROCESSING, 2020, 68 :947-961
[25]   An inertial Bregman generalized alternating direction method of multipliers for nonconvex optimization [J].
Jiawei Xu ;
Miantao Chao .
Journal of Applied Mathematics and Computing, 2022, 68 :1-27
[26]   OPTIMAL COMPUTATIONAL AND STATISTICAL RATES OF CONVERGENCE FOR SPARSE NONCONVEX LEARNING PROBLEMS [J].
Wang, Zhaoran ;
Liu, Han ;
Zhang, Tong .
ANNALS OF STATISTICS, 2014, 42 (06) :2164-2201
[27]   Faster Gradient-Free Algorithms for Nonsmooth Nonconvex Stochastic Optimization [J].
Chen, Lesi ;
Xu, Jing ;
Luo, Luo .
INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 202, 2023, 202
[28]   Interior Epigraph Directions method for nonsmooth and nonconvex optimization via generalized augmented Lagrangian duality [J].
Regina S. Burachik ;
Wilhelm P. Freire ;
C. Yalçın Kaya .
Journal of Global Optimization, 2014, 60 :501-529
[29]   Nonconvex Stochastic Bregman Proximal Gradient Method with Application to Deep Learning [J].
Ding, Kuangyu ;
Li, Jingyang ;
Toh, Kim-Chuan .
JOURNAL OF MACHINE LEARNING RESEARCH, 2025, 26
[30]   Momentum-based variance-reduced stochastic Bregman proximal gradient methods for nonconvex nonsmooth optimization [J].
Liao, Shichen ;
Liu, Yan ;
Han, Congying ;
Guo, Tiande .
EXPERT SYSTEMS WITH APPLICATIONS, 2025, 266