Neural Networks
Backpropagation
10 min+90 XPStep 1/10
Blame assignment, at scale
When a network is wrong, which of its million weights is to blame? Backpropagation answers this with the chain rule: error flows backward through the network, and each weight gets a gradient — its share of responsibility.
Forward pass: make a prediction. Backward pass: assign blame. Update: nudge every weight. That's one training step.
insight
Backprop isn't a different algorithm from gradient descent — it's the efficient way to *compute* the gradient for every weight in a deep network.