A depth dictionary learning method based on a Beta prior process

Abstract

The invention discloses a depth dictionary learning method based on a Beta priori process. The depth dictionary solved by the Beta priori process of the invention is a probability solving method. Unlike arithmetic in solving depth dictionaries, the method of the invention adopts a probability method based on Beta prior process, assumes that the noise, the dictionary and the representation are different Gaussian distributions as prior distributions, adopts a maximum likelihood estimation method to obtain an objective function, solves the objective function by a Gibbs sampling method, and obtains an optimal solution through repeated iterations, thereby effectively denoising. The hierarchical model of depth dictionary can update the posterior distribution of all variables by Gibbs sampling. Since the depth dictionaries are learned directly from sample data, this depth dictionary makes full use of the structural information of the original data. In addition, the variance of reconstructionerror can be learned by nonparametric Bayesian inference.

Description

technical field [0001] The invention relates to the field of deep dictionary learning, in particular to a deep dictionary learning method based on a Beta prior process. Background technique [0002] With the development of artificial intelligence and machine learning, sparse representation dictionary learning is one of the many algorithms of machine learning. Applications. At the same time, machine learning scholars have found that as the number of learning layers deepens, better expression results can be obtained, such as Deep Belief Network (Deep Belief Network) and stacked autoencoder (stacked autoencoder), therefore, deep dictionary Learning came into being. After two years of development, deep dictionary learning has been applied in some aspects, such as cross-modal retrieval, multi-label classification, etc., but not widely; the method of solving deep dictionary is only arithmetic method, such as optimization The minimization algorithm solves the deep dictionary, an...

AI Technical Summary

Problems solved by technology

After two years of development, deep dictionary learning has been applied in some aspects, such as cross-modal retrieval, multi-label classification, etc., but not widely; the method of solving deep dictionary is only arithmetic method, such as optimization The minimization algorithm solves the depth dictionary, and the method of probabilistically solving the depth dictionary has not yet appeared

Images