Title of article :
Agglomerative clustering via maximum incremental path integral
Author/Authors :
Zhang، نويسنده , , Wei and Zhao، نويسنده , , Deli and Wang، نويسنده , , Xiaogang، نويسنده ,
Issue Information :
روزنامه با شماره پیاپی سال 2013
Pages :
10
From page :
3056
To page :
3065
Abstract :
Agglomerative clustering, which iteratively merges small clusters, is commonly used for clustering because it is conceptually simple and produces a hierarchy of clusters. In this paper, we propose a novel graph-structural agglomerative clustering algorithm, where the graph encodes local structures of data. The idea is to define a structural descriptor of clusters on the graph and to assume that two clusters have large affinity if their structural descriptors undergo substantial change when merging them into one cluster. A key insight of this paper to treat a cluster as a dynamical system and its samples as states. Based on that, Path Integral, which has been introduced in statistical mechanics and quantum mechanics, is utilized to measure the stability of a dynamical system. It is proposed as the structural descriptor, and the affinity between two clusters is defined as Incremental Path Integral, which can be computed in a closed-form exact solution, with linear time complexity with respect to the maximum size of clusters. A probabilistic justification of the algorithm based on absorbing random walk is provided. Experimental comparison on toy data and imagery data shows that it achieves considerable improvement over the state-of-the-art clustering algorithms.
Keywords :
Agglomerative clustering , graph algorithms , random walk , path integral
Journal title :
PATTERN RECOGNITION
Serial Year :
2013
Journal title :
PATTERN RECOGNITION
Record number :
1735634
Link To Document :
بازگشت