Multiple graph regularized graph transduction via greedy gradient Max-Cut

被引:9
|
作者
Xiu, Yu [1 ]
Shen, Weiwei [2 ,3 ]
Wang, Zhongqun [1 ]
Liu, Sanmin [1 ]
Wang, Jun [3 ]
机构
[1] Anhui Polytech Univ, Coll Comp & Informat, Wuhu, Anhui, Peoples R China
[2] GE Global Res Ctr, Niskayuna, NY USA
[3] East China Normal Univ, Shanghai Key Lab Trustworthy Comp, Sch Comp Sci & Software Engn, Shanghai, Peoples R China
基金
美国国家科学基金会;
关键词
Graph transduction; Multiple graph regularized; Greedy gradient Max-Cut; Ensemble learning; RECOGNITION; PORTFOLIO; SELECTION; ROBUST;
D O I
10.1016/j.ins.2017.09.054
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
Graph transduction methods have been widely adopted for label prediction under semi supervised settings. To alleviate the relevant sensitivity to initial labels and graph construction processes, recent studies have been aiming at developing robust graph transduction techniques. In particular, the graph transduction method via greedy gradient Max-Cut (GGMC) that minimizes a cost function over a continuous classification function and a binary label variable has been successfully applied to a wide range of applications. However, this method predominately relies on the choice of a high-quality single graph representation, often leading to unstable performance due to selection bias. To tackle this major drawback, we leverage an ensemble learning framework into the GGMC method for exploiting the advantage of constructing and combining multiple graphs. As opposed to performing constrained Max-Cut on a single graph, the proposed multiple graph greedy gradient Max-Cut method (MG-GGMC) simultaneously solves the label prediction and the true graph estimation problems. Specifically, the true graph is approximated by a linear combination of a set of constructed graphs. The coefficients of the linear combination are learned automatically by alternately minimizing a unified objective function in an iterative manner. Comparison studies with representative methods across various real-world benchmarks conspicuously demonstrate the efficaciousness and the superiority of the proposed algorithm in standard evaluation metrics. (C) 2017 Elsevier Inc. All rights reserved.
引用
收藏
页码:187 / 199
页数:13
相关论文
共 50 条
  • [1] Max-Cut Under Graph Constraints
    Lee, Jon
    Nagarajan, Viswanath
    Shen, Xiangkun
    [J]. INTEGER PROGRAMMING AND COMBINATORIAL OPTIMIZATION, IPCO 2016, 2016, 9682 : 50 - 62
  • [2] Approximating graph-constrained max-cut
    Shen, Xiangkun
    Lee, Jon
    Nagarajan, Viswanath
    [J]. MATHEMATICAL PROGRAMMING, 2018, 172 (1-2) : 35 - 58
  • [3] Approximating graph-constrained max-cut
    Xiangkun Shen
    Jon Lee
    Viswanath Nagarajan
    [J]. Mathematical Programming, 2018, 172 : 35 - 58
  • [4] Approximating max-cut under graph-MSO constraints
    Koutecky, Martin
    Lee, Jon
    Nagarajan, Viswanath
    Shen, Xiangkun
    [J]. OPERATIONS RESEARCH LETTERS, 2018, 46 (06) : 592 - 598
  • [5] A MAX-CUT APPROXIMATION USING A GRAPH BASED MBO SCHEME
    Keetch, Blaine
    van Gennip, Yves
    [J]. DISCRETE AND CONTINUOUS DYNAMICAL SYSTEMS-SERIES B, 2019, 24 (11): : 6091 - 6139
  • [6] On greedy construction heuristics for the MAX-CUT problem
    Kahruman, Sera
    Kolotoglu, Elif
    Butenko, Sergiy
    Hicks, Illya V.
    [J]. INTERNATIONAL JOURNAL OF COMPUTATIONAL SCIENCE AND ENGINEERING, 2007, 3 (03) : 211 - 218
  • [7] Experimental performance of graph neural networks on random instances of max-cut
    Yao, Weichi
    Bandeira, Afonso S.
    Villar, Soledad
    [J]. WAVELETS AND SPARSITY XVIII, 2019, 11138
  • [8] Semi-Supervised Learning Using Greedy Max-Cut
    Wang, Jun
    Jebara, Tony
    Chang, Shih-Fu
    [J]. JOURNAL OF MACHINE LEARNING RESEARCH, 2013, 14 : 771 - 800
  • [9] Direction Matters: On Influence-Preserving Graph Summarization and Max-Cut Principle for Directed Graphs
    Xu, Wenkai
    Niu, Gang
    Hyvarinen, Aapo
    Sugiyama, Masashi
    [J]. NEURAL COMPUTATION, 2021, 33 (08) : 2128 - 2162
  • [10] Greedy Differencing Edge-Contraction heuristic for the Max-Cut problem
    Hassin, Refael
    Leshenko, Nikita
    [J]. OPERATIONS RESEARCH LETTERS, 2021, 49 (03) : 320 - 325