If speeds up as grows, and speeds up as grows, how fast does actually grow?
↓ feeds into g(u)
Drag the point on the top curve. Whatever value it lands on, , feeds straight into the bottom curve — the point down there is never dragged directly, it's just wherever sent it.
Call the whole composition . Its derivative is the chain rule —
- — the whole composition, .
- — the inner value, , that feeds into .
- — the outer function's slope, evaluated at the inner value .
- — the inner function's slope with respect to .
- Multiply the two local slopes
In words: the combined rate of change is the outer function's slope (at the inner value ) multiplied by the inner function's slope. Two local slopes, multiplied, give the slope of the whole chain.
↓ feeds into g(u)
Drag the top point and watch both slopes update — then check that their product really does track as you move.
Here and , so .
- Find the inner value at x = 1
- Find both local slopes
- Multiply them (the chain rule)
- Check directly, a different route
Expand the square first: .
Differentiate term by term:
Adding those up: .
Evaluating at : — same answer, two routes.
Drag the top curve’s point until dh/dx reads 20.
↓ feeds into g(u)
To differentiate a composition, multiply the slopes of each stage. This is the entire mechanism behind backpropagation: a network is just a long chain of compositions, and training multiplies slopes all the way back through it.