Unlock instant, AI-driven research and patent intelligence for your innovation.

Training method and device of deep nonlinear principal component analysis network and computer readable storage medium

A non-linear principal component and training method technology, applied in the field of data processing, can solve the problem of high efficiency of data extraction that cannot be complicated, and achieve the effect of strong feature extraction ability

Inactive Publication Date: 2018-06-05
SHENZHEN INST OF ADVANCED TECH
View PDF0 Cites 6 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0006] In view of this, the present invention provides a training method, device and computer-readable storage medium for a deep nonlinear principal component analysis network to solve the problem that the above-mentioned existing nonlinear principal component analysis algorithm cannot be highly effective for complex data extraction characteristic problem

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Training method and device of deep nonlinear principal component analysis network and computer readable storage medium
  • Training method and device of deep nonlinear principal component analysis network and computer readable storage medium
  • Training method and device of deep nonlinear principal component analysis network and computer readable storage medium

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0031] figure 1 It shows the implementation flow of the training method of the deep nonlinear principal component analysis network provided by Embodiment 1 of the present invention. see figure 1 As shown, the implementation process of this method is detailed as follows:

[0032] Step S101, obtaining a deep nonlinear principal component analysis network.

[0033]In this embodiment, the deep nonlinear principal component analysis network acquired in step S101 is an initial deep nonlinear principal component analysis network, and the parameter values ​​in this network are all initial values, and its parameters can be adjusted through training later to obtain the final The deep nonlinear principal component analysis network, after training, the deep nonlinear principal component analysis network can extract highly effective features for complex data.

[0034] Step S102, using the deep non-linear principal component analysis network to perform forward propagation, encode the inp...

Embodiment 2

[0084] Compared with the previous embodiment, the training method of the deep nonlinear principal component analysis network provided by this embodiment uses the deep nonlinear principal component analysis network to perform forward propagation, encodes the input data layer by layer, and solves Before outputting the reconstruction error between the input value of each layer and its estimated value, it also includes:

[0085] Orthogonal constraints are imposed on the k weights in the k-layer deep nonlinear PCA network, and the objective function after the orthogonal constraints is:

[0086]

[0087] Among them, λ is a parameter that controls the strength of orthogonal constraints. If the weight matrix W is limited i Each vector in is a unit vector, then Λ i =I, the initial weight matrix W of the deep nonlinear principal component analysis network under this constraint i A set of basis vectors under the new vector space is formed.

[0088] The goal of deep nonlinear princi...

Embodiment 3

[0094] Compared with Embodiment 1, the training method of the deep nonlinear principal component analysis network provided by this embodiment uses the deep nonlinear principal component analysis network to perform forward propagation, encodes the input data layer by layer, and solves The reconstruction error between the input value of each layer and its estimated value also includes before:

[0095] Sparse constraints are added to the activation values ​​of the hidden layers in the deep nonlinear principal component analysis network, and the objective function of the activation value sparse constraints is:

[0096]

[0097] Among them, β is the coefficient of the activation value sparse term, h(Z i ) is the sparse regularization item of the activation value of the i-th layer, Z i is the activation value of the i-th layer.

[0098] In the deep neural network, we can think that the activation value close to "1" is "active", and the activation value close to "0" is "inactive...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention provides a training method and device of a deep nonlinear principal component analysis network and a computer readable storage medium. The training method comprises the following steps:acquiring the deep nonlinear principal component analysis network; performing forward propagation by using the principal component analysis network for encoding input data layer by layer so as to solve a reconstruction error between an input value and an estimated value of each layer; performing reverse back-transfer from the last layer of the principal component analysis network for calculating agradient of the sum of the reconstruction errors of the whole network for the weight of each layer; according to the gradient of the sum of the reconstruction errors of the whole network for the weight of each layer, calculating a weight matrix of the deep nonlinear principal component analysis network; according to the weight matrix, updating parameters of the principal component analysis network, and returning to the process of performing forward propagation on the input data by using the principal component analysis network until the sum of the reconstruction errors of the whole neural network drops to convergence. According to the invention, highly efficient characteristics can be extracted for complex data.

Description

technical field [0001] The invention belongs to the technical field of data processing, and in particular relates to a training method, device and computer-readable storage medium of a deep nonlinear principal component analysis network. Background technique [0002] Today, with the explosive growth of high-dimensional data, whether in image, video, multimedia processing or in the field of network data correlation analysis, search, biomedical image and bioinformatics, the dimensionality of data has reached thousands or even over The level of 100 million, the number of samples has also reached the same order of magnitude. In the context of high-dimensional and large-scale data, feature extraction and dimensionality reduction are particularly important. Principal component analysis (Principle Component Analysis, PCA) is by far the most widely used dimensionality reduction tool and one of the most important machine learning algorithms. It uses a set of orthogonal transformatio...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06N3/04G06N3/08
CPCG06N3/084G06N3/045
Inventor 乔宇王亚立
Owner SHENZHEN INST OF ADVANCED TECH