DESCENT

Watch a neural net learn.

A tiny network starts knowing nothing — its decision boundary is a random smear. Then gradient descent goes to work: backprop, a loss that falls, a boundary that bends to fit the data. Real backpropagation, hand-written in numpy, no frameworks. Pick a shape and watch it figure out the pattern.

two moons spiral circles xor
Hand-written backprop · numpy · $0 · the model is the point. Part of the Stateful AI galaxy.

↓ descent — watch it learn

class Aclass B moonsspiral circlesxor
EPOCH
LOSS
ACCURACY