keyword
nonconvex support vector machine
A nonconvex support vector machine is a supervised machine learning model for classification or regression whose mathematical formulation results in a nonconvex optimization problem. While standard support vector machines rely on convex surrogate losses like the hinge loss to guarantee a single global minimum, nonconvex support vector machines employ nonconvex loss functions, such as ramp loss, truncated hinge loss, or smooth sigmoidal loss, or nonconvex regularizers. By more closely approximating the ideal discrete zero-one misclassification loss, these models offer increased robustness against outliers and label noise as well as improved feature sparsity, though finding optimal parameters requires specialized optimization techniques capable of handling local minima and stationary points.
1 item
