support vector machine: a standalone construct

Bernhard Boser, Isabelle Guyon, and Vladimir Vapnik introduced the optimal margin classifier in 1992, a method that separates two classes of data by finding the flat boundary, a hyperplane, that sits as far as possible from the nearest examples of either class, so the widest possible gap divides them. Corinna Cortes and Vladimir Vapnik gave the modern form its robustness in 1995 with the support-vector network, adding a soft margin that tolerates a few misclassified or overlapping points so the method still works when the classes are not cleanly separable. The examples that lie on the edge of the gap and pin the boundary in place are the support vectors, and they alone determine the classifier, which is why the method is named for them. A companion idea, the kernel, lets the same margin-finding machinery draw curved boundaries by measuring similarity between examples as if they had been lifted into a higher-dimensional space, without ever computing that space directly. What makes the support vector machine a landmark is that it casts classification as a single convex optimization problem with one objective, the largest margin, which yields a unique solution and a theoretical link between that margin and how well the classifier is expected to generalize to unseen data.[1][2]

Unit

A binary classifier defined by a maximum-margin separating hyperplane and the support vectors that fix it

Recovery means

Not applicable

Measured by

Fit by solving the convex quadratic program that maximizes the margin (or its soft-margin form), with the kernel and the misclassification penalty chosen by cross-validation