Semi-supervised width learning classification method based on manifold regularization and width network

A classification method, semi-supervised technology, applied in neural learning methods, biological neural network models, character and pattern recognition, etc., can solve problems such as limited applicability

Pending Publication Date: 2019-09-27
CIVIL AVIATION UNIV OF CHINA
View PDF0 Cites 12 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] Although BLS are widely used in various fields, they are mainly used in supervised le

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
  • Semi-supervised width learning classification method based on manifold regularization and width network
  • Semi-supervised width learning classification method based on manifold regularization and width network
  • Semi-supervised width learning classification method based on manifold regularization and width network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0142] In Example 1, a series of experiments are carried out on the classic artificial dataset G50C. The G50C dataset is a standard dataset derived from the KEEL database for semi-supervised classification. It is a binary classification dataset in which each class is generated by a 50-D multivariate Gaussian distribution. This classification problem is explicitly designed so that the true Bayes error is about 5%. The dataset consists of 550 samples.

[0143] In order to test the accuracy and efficiency of the proposed method, in this experiment, the G50C data set is divided into 10 labeled training sample sets, 350 unlabeled training sample sets, and 100 labeled test sample sets. The structure of DBN is set to 168-64-32, the number of hidden layer nodes of SS-ELM is set to 1000, the structure of SS-HELM is set to 50-50-500, the structure of SS-BLS is set to 10-10-500, in In this experiment, the parameter selection of the three algorithms is shown in Table 1:

[0144] Table ...

Embodiment 2

[0155] In the second embodiment, a series of experiments are carried out on the classic MNIST handwritten digit images, and the data set consists of 70,000 handwritten digits. Each digit is represented by an image of size 28 × 28 grayscale pixels. Typically, its image is displayed in the Image 6 middle.

[0156] In order to test the performance of the proposed model, in this experiment, the MNIST data set used is 100 labeled training sample sets, 9000 unlabeled training sample sets, and 60000 test sample sets. For reference, the deep structure of DBN is 128-64-32, the structure of SS-HELM is set to 100-100-3000, and the number of hidden layer nodes of SS-ELM is set to 4000. In addition, the parameter selection of DBN is the same as that described in 5.1, and the parameter selection of SS-ELM, SS-BLS, and SS-HELM is shown in Table 3:

[0157] Table 3 Selection of regularization parameters for semi-supervised classifiers on the MNIST dataset

[0158] parameter S...

Embodiment 3

[0172] In the third embodiment, compared with the MNIST data set, the NORB data set is a more complex data set; each image has 2 × 32 × 32 pixels, and there are 48600 images in total. The NORB dataset contains images of 50 different 3D toy objects belonging to 5 different categories: 1) animals; 2) humans; 3) airplanes; 4) trucks; 5) cars, sampled objects are imaged under various lighting conditions and orientations ,Such as Figure 11 shown.

[0173] The training set contains 24300 images of 25 objects (5 of each class), while the test set contains 24300 images of the remaining 25 objects. In the experiment, this paper divides the data into 1000 labeled training sample sets, 14300 unlabeled training sample sets, and 24300 labeled test sample sets. For comparison, the structure of DBN is 128-64-32, SS - The structure of HELM is set to 500-500-3000, and the number of hidden layer nodes of SS-ELM is set to 5000. In addition, the parameter selection of DBN is the same as that ...

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 relates to the technical field of width learning systems, in particular to a semi-supervised width learning classification method based on manifold regularization and a width network, which comprises the following steps of: S1, establishing feature nodes by using input data; s2, establishing an enhanced node by using the established feature node; s3, solving an output weight. According to the invention, the manifold regularization framework is introduced into the BLS; a semi-supervised broad learning system (SS-BLS for short) is proposed, according to the method, BLS is expanded through a manifold regularization framework; therefore, the applicability of the BLS is improved; the SS-BLS completely has the learning capability and the calculation efficiency of the BLS; meanwhile, the method can efficiently complete semi-supervised classification tasks under different complex data sets, the experimental results of various data sets show that SS-BLS has extremely high adaptability and relatively high stability, and in addition, the experimental results of the data sets also prove that the provided method has competitiveness compared with the most advanced semi-supervised method.

Description

technical field [0001] The invention relates to the technical field of width learning systems, in particular to a semi-supervised width learning classification method based on manifold regularization and width networks. Background technique [0002] Deep structure neural network and learning have been applied in many fields, and have achieved great success in large-scale data processing. At present, the most popular deep learning network has deep belief network (Deep Belief Networks, referred to as DBN), deep glass Deep Boltzmann Machines (DBM for short) and Convolutional Neural Networks (CNN for short), etc. Although deep structured networks are very powerful, most networks suffer from an extremely time-consuming training process. The main reason for this is that the deep network structure is relatively complex and needs to adjust a large number of parameters. [0003] In the past few years, scholars from various fields have made significant contributions to the theoretic...

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
IPC IPC(8): G06N3/08G06K9/62
CPCG06N3/08G06F18/241
Inventor 赵慧敏郑建杰邓武徐俊洁
Owner CIVIL AVIATION UNIV OF CHINA
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap