Stochastic Block-Coordinate Gradient Projection Algorithms for Submodular Maximization

We consider a stochastic continuous submodular huge-scale optimization problem, which arises naturally in many applications such as machine learning. Due to high-dimensional data, the computation of the whole gradient vector can become prohibitively expensive. To reduce the complexity and memory req...

Full description

Saved in:
Bibliographic Details
Main Authors: Zhigang Li, Mingchuan Zhang, Junlong Zhu, Ruijuan Zheng, Qikun Zhang, Qingtao Wu
Format: Article
Language:English
Published: Wiley 2018-01-01
Series:Complexity
Online Access:http://dx.doi.org/10.1155/2018/2609471
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849308610869329920
author Zhigang Li
Mingchuan Zhang
Junlong Zhu
Ruijuan Zheng
Qikun Zhang
Qingtao Wu
author_facet Zhigang Li
Mingchuan Zhang
Junlong Zhu
Ruijuan Zheng
Qikun Zhang
Qingtao Wu
author_sort Zhigang Li
collection DOAJ
description We consider a stochastic continuous submodular huge-scale optimization problem, which arises naturally in many applications such as machine learning. Due to high-dimensional data, the computation of the whole gradient vector can become prohibitively expensive. To reduce the complexity and memory requirements, we propose a stochastic block-coordinate gradient projection algorithm for maximizing continuous submodular functions, which chooses a random subset of gradient vector and updates the estimates along the positive gradient direction. We prove that the estimates of all nodes generated by the algorithm converge to some stationary points with probability 1. Moreover, we show that the proposed algorithm achieves the tight (pmin/2F⁎-ϵ) approximation guarantee after O(1/ϵ2) iterations for DR-submodular functions by choosing appropriate step sizes. Furthermore, we also show that the algorithm achieves the tight (γ2/1+γ2pminF⁎-ϵ) approximation guarantee after O(1/ϵ2) iterations for weakly DR-submodular functions with parameter γ by choosing diminishing step sizes.
format Article
id doaj-art-1b372e670d0b4a41b368cb1625d4bff2
institution Kabale University
issn 1076-2787
1099-0526
language English
publishDate 2018-01-01
publisher Wiley
record_format Article
series Complexity
spelling doaj-art-1b372e670d0b4a41b368cb1625d4bff22025-08-20T03:54:24ZengWileyComplexity1076-27871099-05262018-01-01201810.1155/2018/26094712609471Stochastic Block-Coordinate Gradient Projection Algorithms for Submodular MaximizationZhigang Li0Mingchuan Zhang1Junlong Zhu2Ruijuan Zheng3Qikun Zhang4Qingtao Wu5School of Computer and Communication Engineering, Zhengzhou University of Light Industry, Zhengzhou, 450002, ChinaInformation Engineering College, Henan University of Science and Technology, Luoyang, 471023, ChinaInformation Engineering College, Henan University of Science and Technology, Luoyang, 471023, ChinaInformation Engineering College, Henan University of Science and Technology, Luoyang, 471023, ChinaSchool of Computer and Communication Engineering, Zhengzhou University of Light Industry, Zhengzhou, 450002, ChinaInformation Engineering College, Henan University of Science and Technology, Luoyang, 471023, ChinaWe consider a stochastic continuous submodular huge-scale optimization problem, which arises naturally in many applications such as machine learning. Due to high-dimensional data, the computation of the whole gradient vector can become prohibitively expensive. To reduce the complexity and memory requirements, we propose a stochastic block-coordinate gradient projection algorithm for maximizing continuous submodular functions, which chooses a random subset of gradient vector and updates the estimates along the positive gradient direction. We prove that the estimates of all nodes generated by the algorithm converge to some stationary points with probability 1. Moreover, we show that the proposed algorithm achieves the tight (pmin/2F⁎-ϵ) approximation guarantee after O(1/ϵ2) iterations for DR-submodular functions by choosing appropriate step sizes. Furthermore, we also show that the algorithm achieves the tight (γ2/1+γ2pminF⁎-ϵ) approximation guarantee after O(1/ϵ2) iterations for weakly DR-submodular functions with parameter γ by choosing diminishing step sizes.http://dx.doi.org/10.1155/2018/2609471
spellingShingle Zhigang Li
Mingchuan Zhang
Junlong Zhu
Ruijuan Zheng
Qikun Zhang
Qingtao Wu
Stochastic Block-Coordinate Gradient Projection Algorithms for Submodular Maximization
Complexity
title Stochastic Block-Coordinate Gradient Projection Algorithms for Submodular Maximization
title_full Stochastic Block-Coordinate Gradient Projection Algorithms for Submodular Maximization
title_fullStr Stochastic Block-Coordinate Gradient Projection Algorithms for Submodular Maximization
title_full_unstemmed Stochastic Block-Coordinate Gradient Projection Algorithms for Submodular Maximization
title_short Stochastic Block-Coordinate Gradient Projection Algorithms for Submodular Maximization
title_sort stochastic block coordinate gradient projection algorithms for submodular maximization
url http://dx.doi.org/10.1155/2018/2609471
work_keys_str_mv AT zhigangli stochasticblockcoordinategradientprojectionalgorithmsforsubmodularmaximization
AT mingchuanzhang stochasticblockcoordinategradientprojectionalgorithmsforsubmodularmaximization
AT junlongzhu stochasticblockcoordinategradientprojectionalgorithmsforsubmodularmaximization
AT ruijuanzheng stochasticblockcoordinategradientprojectionalgorithmsforsubmodularmaximization
AT qikunzhang stochasticblockcoordinategradientprojectionalgorithmsforsubmodularmaximization
AT qingtaowu stochasticblockcoordinategradientprojectionalgorithmsforsubmodularmaximization