← BACK TO THE LABdy/dx

CALCULUS FOUNDATIONS

From a secant line
to a derivative.

Gradient descent is built on a simple question: if I nudge the input, how much does the output change?

THE LIMIT IN MOTION

Watch the second point approach a

h = 1.00
tangent at a secant across Δx sample points
WHY THIS MATTERS FOR AI

Training follows the slope downhill.

A model changes its parameters in the direction that reduces the loss. Derivatives tell it which small changes help—and by how much.

θ ← θ − α · ∇θ J(θ)