Welsch Loss Based Robust Semi-Supervised Learning

Machine learning involves all aspects of life and achieves good performance,but the training of the traditional fully supervised learning algorithms requires a large deal of manually labeled information.In this case,Semi-Supervised Learning(SSL)is proposed and attracts widespread attention.Due to the limitation of the small number of labeled examples,the performances of existing SSL methods are often affected by the outliers in the labeled data.For the purpose of enhancing the robustness of SSL methods to the outliers,this paper mainly designs two SSL algorithms based on the Welsch loss as well as implements a verification system,which is introduced as follows:In Chapter 2,a robust SSL algorithm based on graph named“Laplacian Welsch Regularization"(LapWR)is proposed.Different from the existing methods,LapWR in-troduces the Welsch loss which can suppress the adverse effect brought by the labeled outliers.To handle the model non-convexity caused by the Welsch loss,an iterative Half-Quadratic(HQ)optimization algorithm is adopted here.Furthermore,this paper proposes an accelerated model based on the Nystrom method.In addition,the gener-alization bound of LapWR is derived theoretically based on analyzing its Rademacher complexity.In Chapter 3,a robust semi-supervised dictionary learning algorithm named "Lapla-cian Welsch Regularization for Robust Semi-Supervised Dictionary Learning"(LWR for short)is proposed.To enhance the robustness of existing methods,this paper introduces the Welsch loss to calculate reconstruction and classification error.Besides,this paper adopt the Laplacian regularizer to enforce similar examples to share similar reconstruction coefficients.In Chapter 4,a robust SSL verification system on the basis of Welsch loss is designed and implemented.This paper first introduces the practical application scenario of the system,and then carries out the system demand analysis.According to the results of system requirement analysis,this paper design and model related modules.Finally,in order to check whether the system module achieves the expected result,a system test analysis is performed.