An animated SVM boundary with two margin lines shows three labeled points with signed margins 1.4, 0.4, and -0.3. The point at 0.4 is correctly classified yet sits inside the margin, while -0.3 is truly misclassified; slack values 0, 0.6, and 1.3 quantify each violation. The hinge loss max(0,1-m) is built and linked to the soft-margin objective, half ||w||² plus C times summed slacks, clarifying C as a regularization-versus-violation tradeoff, applicable even on separable data. Useful for first-year AI students learning SVM formulations.
Narrated · 16:9 · Preview before teaching · automatic layout checks do not establish subject accuracy
# Soft margins penalize more than classification mistakes A classifier's boundary and two margin lines appear around overlapping classes. Introduce signed margins 1.4, 0.4, and -0.3 for three sample cases. Their minimum feasible slack values are 0, 0.6, and 1.3. Show that the sample with margin 0.4 is classified correctly but lies inside the margin, while the sample with margin -0.3 is misclassified. Build the hinge-loss curve max(0,1-m) and connect it to the soft-margin objective: one half of ||w|| squared plus C times the sum of slacks. Explain C as the tradeoff between model regularization and margin violations, without promising that a larger value improves testing performance. Soft margins may also be used on separable data. Keep sum-versus-average loss conventions explicit. Use a black background, white boundaries, and restrained colored violation markers. Useful for distinguishing margin violations from actual prediction errors. Audience: 一年级人工智能专业学生