Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

L0-hinge loss function-based robust classification method

A technology of loss function and classification method, applied in the computer field, can solve the problem that the discreteness of training samples cannot be preserved

Inactive Publication Date: 2018-11-06
SUN YAT SEN UNIV
View PDF0 Cites 3 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The classifiers designed based on these loss functions have certain robustness
However, since this type of loss function has a smoothing effect on misclassified samples, the discreteness of training samples cannot be preserved.

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
  • L0-hinge loss function-based robust classification method
  • L0-hinge loss function-based robust classification method
  • L0-hinge loss function-based robust classification method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0058] The invention discloses a 0 - Robust classification methods for hinge loss functions such as figure 1 As shown, the method steps are as follows:

[0059] S1 obtains training set data

[0060] S2 calculates the kernel matrix K about the training sample, and makes the main diagonal element of the n-dimensional diagonal square matrix Y the training label;

[0061] S3 initializes parameter vectors u and v, penalty function penalty factor β;

[0062] S4 in the robust classifier l 0 - In the SVM model, u and v are solved alternately and iteratively through the block coordinate descent method to minimize the error of the objective function;

[0063] S5 as a robust classifier l 0 -The relative error of the objective function of the SVM model is less than the set threshold, so that the combination coefficient of the hyperplane parameter w Bias b=u n+1 , skip to S6; otherwise, increase penalty factor β, and skip to S4;

[0064] S6 outputs α, b;

[0065] S7 predicts t...

Embodiment 2

[0081] Experiment with real-life binary classification datasets.

[0082] The datasets used in the experiment are provided by the National Taiwan University LIBSVM website, and the datasets used are

[0083] Australian: Australian credit approval data set, a total of 690 samples, each sample has 14 attributes as features;

[0084] breast-cancer: A breast cancer database provided by William H. Wolberg of the University of Wisconsin-Madison Hospital in the United States, with a sample size of 683 and a feature dimension of 10.

[0085] Fourclass: The number of samples is 862, and the feature dimension is 2

[0086] German: German credit data set, which divides the credit risk of people through a set of characteristics, the number of samples is 1000, and the feature dimension is 24.

[0087] Sonar: The number of samples is 208, and the feature dimension of each sample is 60. Of these, 111 samples described sonar signals about metallic objects and 97 samples described sonar sig...

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 present invention provides an L0-hinge loss function-based robust classification method. According to the method, a penalty function and a block coordinate descent method are adopted to solve a classification hyperplane, and predictive judgment is performed on test samples. The method involves a robust classifier L0-SVM model; an L0-hinge loss function has a piecewise constant property, and therefore, the discreteness of misclassified samples will not be smoothed, and distances between the misclassified samples and boundaries do not affect the change of the magnitude of the L0-hinge loss function; the classifier L0-SVM designed based on the L0-hinge loss function is robust to tag noises; and the classification hyperplane with high classification performance can be obtained by means oftraining under a condition that training samples contain tag noises.

Description

technical field [0001] The invention relates to the field of computers, in particular to a 0 - Robust classification methods for hinge loss functions. Background technique [0002] Classification problem is a basic problem in the field of machine learning, and has been successfully applied in many fields, such as computer vision, text classification, medical diagnosis, etc. Classification belongs to supervised learning, and its purpose is to learn a classification model, that is, a classifier, under the supervision of which class each training sample is determined to belong to. This model can map the data items in the database to a given class. category. [0003] Among them, support vector machine (SVM) is applied to classification problems as a supervised learning method. For unpolluted training samples, SVM usually has good classification performance, but for training samples polluted by label noise, The performance of the support vector machine may be weakened, because...

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): G06K9/62
CPCG06F18/2411
Inventor 李洽唐建雄
Owner SUN YAT SEN UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com