论文部分内容阅读
Support vector machines(SVMs) have been intensively applied in the domains of speech recognition, text categorization, and faults detection. However, the practical application of SVMs is limited by the non-smooth feature of objective function. To overcome this problem, a novel smooth function based on the geometry of circle tangent is constructed. It smoothes the non-differentiable term of unconstrained SVM, and also proposes a circle tangent smooth SVM(CTSSVM). Compared with other smooth approaching functions, its smooth precision had an obvious improvement. Theoretical analysis proved the global convergence of CTSSVM. Numerical experiments and comparisons showed CTSSVM had better classification and learning efficiency than competitive baselines.
However, the practical application of SVMs is limited by the non-smooth feature of objective function. To overcome this problem, a novel smooth function based on the geometry of circle tangent is constructed. It smoothes the non-differentiable term of unconstrained SVM, and also proposes a circle tangent smooth SVM (CTSSVM). Compared with other smooth approaching functions, whose smooth precision had an obvious improvement. Theoretical analysis proved the global convergence of CTSSVM. Numerical experiments and comparisons showed showed CTSSVM had better classification and learning efficiency than competitive baselines.