A feature learning model based on adaptive dropout non-negative matrix factorization

Abstract

The present invention provides an adaptive Dropout non-negative matrix factorization based feature learning model based on an analysis on a relationship between hidden features in the NMF. A dissimilarity degree between the hidden features is capable of being studied actively, and can be converted into data representation ability of the hidden features. Then a probability function is built on the above basis, and the Dropout is performed on the hidden features, thus an interaction effect of the hidden features in an optimizing process is reduced, and semantic independence of the hidden features is improved. The invention has the advantages of being good in interpretability and generalization, and can acquire significant performance improvement on texts and image data, and can be applied to the existing algorithm based on the NMF. Besides, the adaptive Dropout non-negative matrix factorization based feature learning model also has the advantage of being good in parallelizability, and can be deployed on a parallel platform for handling large-scale data.

Description

technical field [0001] The invention belongs to the technical field of computer applications, in particular to data mining and machine learning, in particular to a feature learning model based on adaptive Dropout non-negative matrix decomposition. Background technique [0002] With the development of Internet technology and the rise of social networks, the means of obtaining and sharing information are becoming more and more convenient. The Internet is full of unstructured data such as texts and images. At the same time, due to the arbitrariness and irregularity of data release, there is a lot of noise in the data. Even after data cleaning, the data still faces problems such as data sparseness and high dimensionality. Therefore, feature learning is often required before clustering, classification, recommendation and other tasks. [0003] Non-negative Matrix Factorization (NMF) is a more popular multivariate analysis model in recent years. Because of its good interpretabili...

AI Technical Summary

Problems solved by technology

[0008] The purpose of the present invention is to solve the problem of semantic ambiguity and semantic overlap of hidden features existing in the existing NMF model, and to provide a feature learning model based on adaptive Dropout non-negative matrix decomposition

Images