Research On Joint-regularization Based On Semi-supervised Classification Method

In the age of Big Data,it is often not practical for man to process massive amounts of data by hand.For this reason,machine learning has been created using computers as a tool and combining statistics and other disciplines.Machine learning methods provide a solution for computers to process massive amounts of information.In many practical problems,it is necessary to classify samples.Usually,there are fewer samples of class labels and more samples without class labels.Due to factors such as time or cost,it is difficult to mark all samples in a short time.Therefore,in order to solve the problem of class labels,a large number of unlabeled samples are added on the basis of supervised learning,and the model is jointly trained to predict class labels of unlabeled samples,thus resulting in semi-supervised learning.In recent years,the ideas and methods of semi-supervised learning have been widely used in many fields such as engineering,biology,medical care,and finance.At present,the study of semi-supervised learning mainly focuses on the regularization framework of manifolds,that is,constructing the regularization of manifolds to measure the geometric structure of samples.The existing model improvements mainly include:introducing pairs of constraints and other methods;improving the loss function,such as adding a projection method to the loss function;changing the structure of the model,such as extending the support vector machine model to twins support vector machine model;introduction of relevant criteria,such as introducing the maximum correlation entropy in information theory to improve the robustness of the model.In this paper,two semi-supervised models with joint regularization were proposed based on the related research.In order to obtain more empirical information,pairwise constraints and maximum correlation entropy are introduced in the framework of manifold regularization.A semi-supervised classification model based on the maximum correlation entropy criterion is proposed and the parameter estimation of the model is given.The pairwise constraints item is added to the model of the semi-supervised project twins support vector machine model,A semi-supervised project twins support vector machine model based on pairwise constraints is proposed,and the parameter estimation of the model is given.Atthe same time,starting from the semi-supervised classification method,the semi-supervised classification model in different contexts is described and compared.Combined with the generative method,the general steps of the semi-supervised generative method are summarized.Finally,the classification accuracy of the proposed model and the existing model is compared on the constructed data set and UCI data set respectively.Experimental results show that the improved semi-supervised model improves classification accuracy and model robustness to some extent.