Adaptive graph constraint multi-view linear discriminant analysis method and system and storage medium

Abstract

The invention discloses an adaptive graph constraint multi-view linear discriminant analysis method and system and a storage medium. The method comprises the following steps: acquiring common low-dimensional representation of multiple views and a projection matrix of each view; a graph capable of reflecting internal distribution of data is adaptively learned through public low-dimensional representation to constrain the learning process; learning an optimal projection matrix by using a linear discrimination method; and performing feature extraction on the test set multi-view data through the optimal projection matrix, and inputting a feature extraction result into a KNN classification system to complete a classification task. According to the method, multi-view canonical correlation analysis and linear discriminant analysis are combined and uniformly expressed, a multi-view learning feature extraction method is more suitable for classification tasks than a single-view feature extraction method, label information of unknown label data is utilized in linear discriminant analysis, common source knowledge induced by graphs is considered at the same time, and the classification efficiency is improved. The distance between expected standard variables is minimized, and fusion extraction and dimension reduction of multi-view data features are realized.

Description

technical field [0001] The invention belongs to the technical field of multi-view data classification, and in particular relates to an adaptive graph-constrained multi-view linear discriminant analysis method, system and storage medium. Background technique [0002] In many practical applications, multi-view learning is usually more robust than single-view learning. A key issue in multi-view learning is how to effectively utilize the information from different feature sets. An efficient method is to fuse the information by acquiring a common subspace for these feature sets, which is usually implemented using feature extraction. Canonical correlation analysis is a classic tool for multi-view learning, which learns a linear projection matrix for each view separately by maximizing the cross-correlation between the two views. Multi-view canonical correlation analysis is an extension of canonical correlation analysis on multi-view data, which can be used to process data with mor...

AI Technical Summary

Problems solved by technology

Such a method does not use the label information of the data in the first step of solving the common subspace, which may lead to the learned common subspace is not the optimal solution

[0006] (3) Many existing multi-view feature extraction algorithms are only effective on data sets with a small number of views and a small data size. When the number of views is large or the data size is large, it will be due to too many hyperparameters or other reasons lead to excessive computational complexity, making it difficult to process these large datasets

Images