ML Interactive Lab – SVM & Gradient Descent

This lab connects two core ideas: how an optimizer moves on a loss landscape, and how a classifier shapes a decision boundary. The SVM tab shows soft-margin geometry, kernels, and KKT-style α-bars. The Gradient Descent tab shows how parameters slide down a quadratic bowl while the loss curve drops over iterations.

C (Slack Penalty): 1.0

Training Strength (epochs): 25

Kernel: Background shows sign of f(x) in input space; non-linear shapes appear for Poly/RBF.

Learning Rate η: 0.050 Left: gradient descent path on a 2D quadratic bowl. Right: the matching loss J(w) vs iteration. Try increasing η to see overshooting or divergence.