Support-Vector Networks

Summary

Introduces the support vector machine with a soft margin, separating classes by the widest possible margin while permitting bounded violations, and using kernels to obtain non-linear boundaries.

Why it matters

It combined a clear geometric principle, maximize the margin, with a convex optimization that has a unique solution, and the kernel trick let the same machinery produce non-linear boundaries without constructing high-dimensional features. SVMs dominated practical classification until deep learning displaced them on perceptual tasks.