How Models Learn

Backpropagation

The method for working out how much each weight contributed to the loss, by passing errors backwards through the network.

Backpropagation is the bookkeeping trick that finds this direction for every weight at once, by passing the blame for each mistake backwards through the layers.
From Training a network, step “Loss and backpropagation”

See it in action

Try backpropagation hands-on in the lesson Training a network, part of the AI fundamentals course.

Open the lesson