Improved spectral clustering and parallelization method

Abstract

The present invention discloses an improved spectral clustering method based on a swarm intelligence algorithm. Feature vectors corresponding to former 2k maximum feature values of a Laplacian matrixare selected as source data of a cluster, and the swarm intelligence algorithm is employed to select high-quality initialization center points for cluster operation so as to improve the maximum accuracy of the cluster and the stability of multi-times cluster results. The Cuckoo search algorithm is introduced to search initialization center points, a fitness function in the process of the Cuckoo search algorithm employs an error square sum function and is applied into the spectral clustering to take data points of the minimum error sum square obtained through search as the initialization centerpoints. The Levy flight strategy in the Cuckoo search algorithm is introduced into the particle swarm optimization algorithm, in a condition that the convergence rate of the particle swarm optimization algorithm is slow, the Levy flight strategy is employed to generate step lengths with small frequencies and occasionally large step lengths, and speed updating formulas for different aspects are introduced in different step lengths.

Description

technical field [0001] The invention belongs to a non-supervised learning method in machine learning, and relates to a clustering method and a group intelligence algorithm. Background technique [0002] The purpose of clustering is to divide the data, divide similar data into the same cluster, and divide dissimilar data into different clusters. With the development of information technology, the diversity of data has been brought, which makes the dimensions of many data irrelevant. It is difficult for traditional clustering algorithms to deal with these uncorrelated data. Spectral clustering is a new type of clustering algorithm, which can cluster in sample space of arbitrary shape and can converge to the global optimal solution, and has been widely used in computer vision, text mining and biological information mining and other fields. At present, the research on spectral clustering mainly includes the construction of similarity matrix, the selection of eigenvectors, the d...

AI Technical Summary

Problems solved by technology

With the development of information technology, the diversity of data has been brought, which makes the dimensions of many data irrelevant. It is difficult for traditional clustering algorithms to deal with these unrelated data.

Images