Research About Multi-view Classifier Based On Robust Representation Of View Knowledge

Nowadays,many artificial intelligence technologies have been successfully applied to all aspects of people’s daily lives.As a typical research object in machine learning,classifiers have an essential position in artificial intelligence theoretical research.Many classification algorithms have been proposed,and the Support Vector Machine(SVM: Support Vector Machine)is one of its classic methods.In the era of big data,multi-view data describing the same target in different types is spread across various application fields,such as photos taken of the same person from multiple angles.Taking multi-view data as the research object,some scholars have proposed an algorithm called Multiview Generalized Eigenvalue Proximal Support Vector Machines(Mv GSVMs)in recent years.This algorithm improves the model’s classification accuracy through the organic integration of multi-view learning and classification tasks.Nevertheless,the algorithm still suffers from many problems.For example,the model’s robustness cannot be guaranteed,and the ability to mine the potential relationships between views is lacking.In order to overcome these problems,based on Mv GSVMs,this paper proposes a series of effective multi-view support vector machine methods.This paper examines the effectiveness of these algorithms on UCI,face image,forest fire,and other databases.The innovative results of this paper mainly include:(1).Propose an Improved multi-view GEPSVM(IMv GEPSVM)algorithm.Unlike the existing multi-view SVM method,IMv GEPSVM introduces the constraint of maximizing the difference of different views into the model,thereby improving the generalization ability of the model.In order to improve the model’s robustness,using L1-norm to measure the distance from the point to the plane in this model.Simultaneously,this paper proposed an effective iterative solution to solve the difficulties caused by the L1-norm distance measurement and the difference about the regular term of different views in the objective function.From the theoretical level,the convergence of the algorithm is proved.The feasibility of IMv GEPSVM is verified on UCI,face images,and forest fire data sets.(2).In order to dig out the potential relationship between views,this paper proposed an algorithm that is based on the consensus principle of multi-view learning——Multi-view Generalized Support Vector Machine via Mining the Inherent Relationship between Views(MRMv GSVM).Different from the calculation of the distance between the sample point and the hyperplane that indicates the consistency between different views of the same class,this paper uses a defined relationship matrix to connect the two views.At the same time,in order to select the most relevant sample data from data samples to establish a multi-view common regular term,this paper uses the L2,1-norm constraint relationship matrix.This paper proves the effectiveness of the algorithm on UCI,face and firework image datasets.