A fast spectral clustering method based on improved kd-tree marker selection

Abstract

The invention discloses a fast spectrum clustering method based on improved kd tree mark point selection. The invention firstly uses the improved kd tree method to select the marker points in the dataset, then establishes the similarity matrix between the data according to the marker points, and finally uses a depth self-coding instead of the traditional dimension reduction method for eigendecomposing the Laplace matrix. In the experiment, we prove the superiority of our method. In the large-scale data set, the algorithm improves the speed of clustering greatly with the accuracy basically consistent with the classical spectral clustering algorithm.

Description

technical field [0001] The invention relates to the field of clustering under data mining, in particular to a fast spectral clustering method based on the selection of improved kd tree marker points. Background technique [0002] Spectral clustering comes from the minimum cut problem in graph theory. If the data set is represented as a graph, each data point is used as a vertex of the graph, and the edges connecting the vertices represent the similarity between adjacent vertices. Then the process of spectral clustering is to divide the similarity graph into several subgraphs by cutting the edge with the smallest similarity, so that the similarity in the subgraph is the largest and the similarity between the subgraphs is the smallest. [0003] Spectral clustering works very well for high-dimensional small-scale data clustering because of its unique step of decomposing the Laplacian matrix into features. However, the computational complexity of this eigen-decomposition step ...

AI Technical Summary

Problems solved by technology

But the size of the input layer of the neural network is equal to the number of data points, as n grows, the entire network will expand dramatically

Images