Gradient Descent & Cost Functions — Interactive Explorer

Animate optimization on 2D loss landscapes — tune learning rate, compare GD variants, discover saddle points and flat regions

Learning Rate η

η 0.100

moderate — steady convergence

Gradient Descent Variant

exact gradient — smooth deterministic path

Simulation

Click the landscape to reposition the ball

Cost Function

—

What's happening?

Select a landscape and press Play.

Step 0

Loss —

‖∇L‖ —

Loss Landscape (w₁, w₂) — contours & optimization path

click to set start · dragging resets

Low → High loss Ball ★ Global min

Loss Curve — iteration vs. L(w)