Semi-Supervised Learning in Network-Structured Data via Total Variation Minimization

被引:27
作者
Jung, Alexander [1 ]
Hero, Alfred O., III [2 ]
Mara, Alexandru Cristian [3 ]
Jahromi, Saeed [4 ]
Heimowitz, Ayelet [5 ]
Eldar, Yonina C. [6 ]
机构
[1] Aalto Univ, Dept Comp Sci, Espoo 02150, Finland
[2] Univ Michigan, Dept Elect Engn & Comp Sci, Ann Arbor, MI 48109 USA
[3] Univ Ghent, Dept Elect & Informat Syst, B-9052 Ghent, Belgium
[4] Monash Univ, Fac Informat Technol, Melbourne, Vic 3800, Australia
[5] Princeton Univ, Program Appl & Computat Math, Princeton, NJ 08544 USA
[6] Weizmann Inst Sci, Fac Math & Comp Sci, IL-7610001 Rehovot, Israel
关键词
Machine learning; semisupervised learning; optimization; big data applications; network theory (graphs); FLOW; RECOVERY; SIGNALS;
D O I
10.1109/TSP.2019.2953593
中图分类号
TM [电工技术]; TN [电子技术、通信技术];
学科分类号
0808 ; 0809 ;
摘要
We provide an analysis and interpretation of total variation (TV) minimization for semi-supervised learning from partially-labeled network-structured data. Our approach exploits an intrinsic duality between TV minimization and network flow problems. In particular, we use Fenchel duality to establish a precise equivalence of TV minimization and a minimum cost flow problem. This provides a link between modern convex optimization methods for non-smooth Lasso-type problems and maximum flow algorithms. We show how a primal-dual method for TV minimization can be interpreted as distributed network optimization. Moreover, we derive a condition on the network structure and available label information that ensures that TV minimization accurately learns (approximately) piece-wise constant graph signals. This condition depends on the existence of sufficiently large network flows between labeled data points. We verify our analysis in numerical experiments.
引用
收藏
页码:6256 / 6269
页数:14
相关论文
共 57 条
  • [1] Abbe E, 2018, J MACH LEARN RES, V18
  • [2] Amdahl G. M., 1967, P APR 18 20 1967 SPR, P483, DOI DOI 10.1145/1465482.1465560
  • [3] Efficient Sampling Set Selection for Bandlimited Graph Signals Using Graph Spectral Proxies
    Anis, Aamir
    Gadde, Akshay
    Ortega, Antonio
    [J]. IEEE TRANSACTIONS ON SIGNAL PROCESSING, 2016, 64 (14) : 3775 - 3789
  • [4] [Anonymous], 2014, Found. Trends Optim.., DOI DOI 10.1561/2400000003
  • [5] [Anonymous], 2016, Network Science
  • [6] [Anonymous], 2013, Graphs, Networks and Algorithms
  • [7] Bauschke HH, 2011, CMS BOOKS MATH, P1, DOI 10.1007/978-1-4419-9467-7
  • [8] NESTA: A Fast and Accurate First-Order Method for Sparse Recovery
    Becker, Stephen
    Bobin, Jerome
    Candes, Emmanuel J.
    [J]. SIAM JOURNAL ON IMAGING SCIENCES, 2011, 4 (01): : 1 - 39
  • [9] Regularization and semi-supervised learning on large graphs
    Belkin, M
    Matveeva, I
    Niyogi, P
    [J]. LEARNING THEORY, PROCEEDINGS, 2004, 3120 : 624 - 638
  • [10] Bertsekas D. P., 1998, Network optimization: continuous and discrete models