Gradient
Lesson 4.3MathCode~50 min

The chain rule

Gears turning gears: how change flows through a chain of functions. (Backprop in disguise.)

Module 4 · 0/7 done

Functions inside functions

Real models aren’t one simple function. They’re chains: the input goes through one function, whose output goes into another, and another. So we need to answer: if y=f(g(x))y = f(g(x)), how fast does yy change when we nudge xx?

Think of gears. If gear x turns gear u three times as fast, and u turns gear y half as fast as itself, then y turns 3×12=1.53 \times \tfrac{1}{2} = 1.5 times as fast as x. The rates multiply.

xuyx turns →u turns faster →y turns too.Rates multiply.
Fig 4.4Linked gears: each one’s speed is the previous speed times a ratio. Derivatives chain the same way.