Derivatives, gradients, the chain rule, and gradient descent built by hand before autograd does it for you.
12h · 9 lessons · 1 challenge
Before this: Linear Algebra for Machine Learning
Backpropagation is the chain rule applied to a graph, and nothing more. This module makes that literal: you differentiate scalar functions, then vector functions, then a two-layer network on paper, and only then write the autograd engine that does it for you.
The optimization half covers why gradient descent works, why it stalls, and what momentum, RMSProp and Adam each fix — so that when a loss curve goes flat you have a hypothesis rather than a shrug.
A network is drawn as a graph. Propagate the gradient node by node against the clock; your answer is checked against autograd to four decimal places.
Worth 450 XP.
