A Clustering Method Based on Deep Neural Networks and Pairwise Constraints

Abstract

The invention discloses a clustering method based on a deep neural network and pairwise constraints. A data set containing pairwise constraints between data is given; a difference vector between samples of the data set is obtained; an autoencoder network and a deep neural network are constructed. Network; the data set sample is used as the input of the self-encoding network, the input data set sample is used as the output of the self-encoding network to train the network, the output of the bottleneck of the self-encoding network is used as the input of the deep neural network, and the pairwise constraints are used as the correct label training network ; Combine trained autoencoder network and deep neural network to clustering algorithm; use clustering algorithm for clustering tasks. The invention combines the pairwise constraints between the data in the original data set, performs dimensionality reduction operations on the input data and deep neural network learning features through the autoencoder network, and proposes the loss function of the network model and its optimization algorithm based on gradient descent. Effectively improve the clustering accuracy of the clustering algorithm.

Description

technical field [0001] The invention relates to the technical field of clustering methods based on deep neural networks and pair constraints and high-dimensional clustering, in particular to a method for clustering based on pair constraints between data. Background technique [0002] Data clustering, also known as unsupervised learning, is an efficient method for dividing a set of data objects into several clusters. But unsupervised learning cannot know what each cluster represents, because it clusters unlabeled data. With the continuous deepening of network informatization, the total amount of data on the entire Internet is constantly increasing. How to fully explore and utilize the useful information contained in the data has become a hot issue in the field of computer science in recent years. High-dimensional clustering is a common problem. The specific performance is that traditional clustering algorithms encounter difficulties in clustering high-dimensional data spaces...

AI Technical Summary

Problems solved by technology

However, the current use of neural networks for dimensionality reduction operations is to directly learn the characteristics of the original data, while ignoring the pairwise constraints between the original data.

[0004] Since most of the existing high-dimensional clustering methods using neural networks have the disadvantage of directly learning the characteristics of the original data, even if a large amount of data is used to train the neural network, the accuracy of clustering cannot be further improved

Images