If \(y = f(g(h(x)))\), differentiate step by step:
\[ \frac{dy}{dx} = f'(g(h(x))) \cdot g'(h(x)) \cdot h'(x) \]Example: \(y = \sin(3x^2 + 1)\)
Outer: \(\sin(u),\, u = 3x^2 + 1\)
\(dy/du = \cos(u),\, du/dx = 6x\)
\(dy/dx = \cos(3x^2 + 1) ยท 6x\)
\(y = e^{\sin(2x)}\)
Let \(v = \sin(2x) โ y = e^v\)
\(dy/dv = e^v,\, dv/dx = \cos(2x) ยท 2\)
\(dy/dx = e^{\sin(2x)} ยท 2 \cos(2x)\)
Common in ML: activation functions like sigmoid, ReLU derivatives.