Towards a theoretical foundation for Laplacian-based manifold methods

被引:124
作者
Belkin, M [1 ]
Niyogi, P [1 ]
机构
[1] Univ Chicago, Dept Comp Sci, Chicago, IL 60637 USA
来源
LEARNING THEORY, PROCEEDINGS | 2005年 / 3559卷
关键词
D O I
10.1007/11503415_33
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
In recent years manifold methods have attracted a considerable amount of attention in machine learning. However most algorithms in that class may be termed "manifold-motivated" as they lack any explicit theoretical guarantees. In this paper we take a step towards closing the gap between theory and practice for a class of Laplacian-based manifold methods. We show that under certain conditions the graph Laplacian of a point cloud converges to the Laplace-Beltrami operator on the underlying manifold. Theorem 1 contains the first result showing convergence of a random graph Laplacian to manifold Laplacian in the machine learning context.
引用
收藏
页码:486 / 500
页数:15
相关论文
共 30 条
  • [21] NIYOGI P, 2004, TR200408 U CHIC
  • [22] Rosenberg S., 1997, LAPLACIAN RIEMANNIAN, DOI DOI 10.1017/CBO9780511623783
  • [23] Nonlinear dimensionality reduction by locally linear embedding
    Roweis, ST
    Saul, LK
    [J]. SCIENCE, 2000, 290 (5500) : 2323 - +
  • [24] Normalized cuts and image segmentation
    Shi, JB
    Malik, J
    [J]. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 2000, 22 (08) : 888 - 905
  • [25] SMOLA A, 2003, KERNELS REGULATIZATI
  • [26] SPIELMAN D, 1996, SPECTRAL PARTITIONIN
  • [27] SZUMMER M, 2001, PARTIALLY LABELED CL
  • [28] A global geometric framework for nonlinear dimensionality reduction
    Tenenbaum, JB
    de Silva, V
    Langford, JC
    [J]. SCIENCE, 2000, 290 (5500) : 2319 - +
  • [29] VONLUXBURG U, 2004, 134 M PLANCK I BIOL
  • [30] ZOMORODIAN A, 2004, 20 ACM S COMP GEOM