Many lines separate two classes — which is best? The SVM picks the one with the widest empty margin on both sides: the fattest ‘street’ between the classes. Only the closest points — the SUPPORT VECTORS — touch the curb and define the boundary; the rest don’t matter. Maximizing that gap gives the most robust generalization. Slide to reveal the margin.
A Support Vector Machine finds the separating hyperplane that MAXIMIZES THE MARGIN — the distance to the nearest points of each class. Equivalently it minimizes ½||w||² subject to yᵢ(w·xᵢ+b) ≥ 1, so the margin is 2/||w||. Only the closest points (the SUPPORT VECTORS) are active constraints and determine the boundary; all others are irrelevant. Max-margin gives strong generalization; the kernel trick lets it draw non-linear boundaries in a lifted space, and a soft margin tolerates overlap. A fail-loud self-check throws unless the max-margin line separates the two clusters with the expected margin. ◆ real machine learning, node-verified.
The hard-margin, linearly-separable case is shown; real data uses soft margins (slack) and kernels. The max-margin objective, the 2/||w|| width, and support-vectors-only dependence are exact.