Probability pairwise constraint-based self adaptive semi-supervised dimensionality reduction method

Abstract

The invention discloses a probability pairwise constraint-based self adaptive semi-supervised dimensionality reduction method comprising the following the steps: a projection matrix is calculated, a sparse representation graph is constructed, and weight of pairwise constraints can be calculated; according to an amount of information contained in each constraint, differentiating treatment can be performed in a self adaptive manner, a sparse representation method is used for capturing an inner structure of data, and a quadratic programming algorithm is adopted for assessing probabilities accounted for by information of all constraints; calculation of the projection matrix, construction of a adjacent graphs and updating of constraint weight are combined organically so as to form an integral body; the projection matrix is finally obtained, and low dimension representation of data can be obtained after the projection matrix is applied to original data. The method disclosed in the inventionis simple in required supervision information, high in calculation speed and large in popularization value; pertinent dimensionality reduction can be realized.

Description

technical field [0001] The invention relates to the technical field of semi-supervised machine learning, in particular to an adaptive semi-supervised dimensionality reduction method based on probability pair constraints. Background technique [0002] In the fields of computer vision, speech recognition, natural language processing, and bioinformatics, the data that needs to be processed by computers often has extremely high dimensions, but such high-dimensional data has brought great trouble to the practical application of machine learning methods. For example, in a classification problem, when the number of samples is much smaller than the dimension of the sample, this high dimensionality will directly lead to a decrease in the performance of the classifier, which is the so-called "curse of dimensionality" problem. However, it is found that many high-dimensional data are mapped by low-dimensional manifolds, which makes researchers look for various dimensionality reduction m...

AI Technical Summary

Problems solved by technology

Therefore, the quality of the dimensionality reduction results largely depends on whether the constructed neighborhood map is appropriate, and the construction of the neighborhood map itself is an unsupervised process, so the obtained neighborhood map may not reflect the user's real Intent, it cannot effectively guide the dimensionality reduction process

[0008] Existing dimensionality reduction methods have at least these problems: the purpose of dimensionality reduction is not clear (unsupervised dimensionality reduction); either requires data labels, which are usually difficult to obtain (supervised or semi-supervised dimensionality reduction); treat all constraints equally Information (pair-constrained semi-supervised dimensionality reduction); the construction of the graph is separated from the dimensionality reduction step, and it cannot be adaptive to deal with different data, etc.

These problems will bring great inconvenience to subsequent applications.

Images