Backpropagation is how neural networks are trained

For every training example, we compute loss function when we finish, then adjust each weight based on how much it contributed to correct answer