Robust classifier learning with fuzzy class labels for large-margin support vector machines

摘要: Using class label fuzzification, this study develops the idea of refreshing attitude difficult training examples and gaining a more robust classifier for large-margin support vector machines (SVMs). Fuzzification relaxes specific hard-limited Lagrangian constraints examples, extends infeasible space canonical optimization, reconfigures consequent decision function with wider margin. With margin, capable achieving high generalization performance can be robust. This paper traces rationale such back to changes governing loss function. From aspect function, reasons are causally explained. In study, we also demonstrate two-stage system experiments show corresponding fuzzification. The first captures in first-stage preprocessor, assigns them various fuzzified labels. Three types membership functions, including constant, linear, sigmoidal designated preprocessor manipulate within-class correlations reference benchmarks confirm generalized ability due Since change y"i^' is fundamental, may transplanted different prototypes SVM.