Method for quickly solving sparse codes by using hybrid sparse neural network

Abstract

The invention discloses a method for quickly solving sparse codes by using a hybrid sparse neural network. The method comprises the steps of S1 adopting a rough estimation neural network, the hybrid sparse neural network and a sampling module to jointly serve as a network solving model; S2 inputting the input data into a coarse estimation neural network for prediction to obtain an initial prediction result of sparse coding; S3 inputting the initial prediction result into the hybrid sparse neural network for prediction to obtain a probability distribution parameter of sparse coding; and S4 inputting the probability distribution parameter of sparse coding into a sampling module to sample a probability distribution result, and obtaining sparse coding according to the sampling result. According to the method, because probability distribution prediction and sparse coding solving modes are combined, compared with an existing sparse coding solving method, probability distribution of prediction coding can better describe coding, and sparse coding can be accurately solved.

Description

technical field [0001] The invention relates to the field of artificial intelligence, in particular to a method for quickly solving sparse coding by using a mixed sparse neural network. Background technique [0002] Sparse coding has proven to be a great success in the field of recovering semantic information of high-dimensional data. Therefore, sparse coding has become very popular in extracting features of raw data, and its applications include: image super-resolution, subspace learning, image classification and a series of fields. [0003] In order to obtain the sparse coding of the original data, traditional methods mainly use high-dimensional convex optimization algorithms to solve them, such as the Iterative Shrinkage Threshold (ISTA) algorithm, but the computational complexity is very high. To this end, the researchers expanded the traditional iterative shrinkage threshold calculation module into a recurrent neural network to improve the calculation speed, such as th...

AI Technical Summary

Problems solved by technology

[0004] Based on the problems existing in the prior art, the purpose of the present invention is to provide a method for quickly solving sparse coding using a hybrid sparse neural network, which can solve the problem of inaccurate estimated values ​​of sparse coding in existing methods for solving sparse coding based on neural networks. question

Images