Lesson 7.6Model~1 h

Training a network

A live playground: watch a network bend its decision boundary around a spiral.

Module 7 · 0/11 done

A network playground

Everything comes together here: a real neural network, trained live with backpropagation. Each small square is one neuron, showing what it responds to across the whole input plane. The big square is the network’s final answer, with training points filled and test points hollow.

Fig 7.6Press Train. Then experiment: switch to the spiral, remove the hidden layers entirely, add neurons, change the activation, crank the learning rate. Watch the hidden neurons — each one learns a different ‘feature’.