How a network computes, why gradients vanish, and what makes depth trainable
The backward pass has to start somewhere. Every gradient in a network is the loss's derivative pushed back through one layer, then the next, then the next. Before any of that can happen you need the very first number: how much does the loss change when the prediction changes? That number is ∂L/∂ŷ, and it is where backpropagation begins.
🔒 This is a Pro lesson — the interactive figure, worked examples, quiz and practice open with Pro access.
▶ Where Backprop Starts: the Derivative of the Loss