Multi-output Support Vector Regression Machines And ?-regularized Extreme Learning Machines

This paper mainly studied the various forms of support vector regression machines(SVRs)and extreme learning machines(ELMs),and we make a comparison about the superiority and inferiority in dealing with multiple output problems.This paper mainly divided into four chapters.The first chapter briefly describes the various forms of SVRs model and the models of the ELMs.In order to improve the non sparsity of the regularized extreme machine learning(RELM)and the regularized kernal extreme machine learning(RKELM),?-RELM and ?-RKELM were proposed based on the Vapnik ?-insensitive loss function and the effectiveness of the proposed algorithm is verified by experiments.They have the advantage in two aspects,on the one hand,they have more fast calculation speed;On the other hand the calculation precision is very high.In the third and fourth chapters introduce the mathematical model of multiple output SVR(M-SVR)types and the multiple output twin SVR(M-TSVR)respectively,and they are the promotion and improvement for the single output SVR and single output twin SVR.For this method,the maximum advantage is considered the correlation between the different output.