Backpropagation: how a neural net’s loss gradient flows backward through every parameter. The single most important algorithm to write yourself before relying on a framework.
[ ] Not started.
prereqs/calculus/ — chain rule, partial derivatives.
micrograd — best entry point.