Layers of simple units, each weighing its inputs, learn by nudging weights after every mistake.
- NeuronMultiply inputs by weights, add them, pass through an activation.
- LayersStack neurons so later layers build on earlier ones.
- Forward passData flows through to a prediction.
- BackpropSend the error backwards to work out each weight's blame.
- UpdateNudge every weight a little and repeat.
Explain 1/6
Source: PyTorch: Build the neural network