Backpropagation
Also called Backpropagation
The algorithm that figures out how much each weight contributed to the error, so it can be corrected.
Think of it like
Like tracing a factory defect back through the assembly line to see which station caused it, and by how much.
Example
After a wrong prediction, backprop pushes the error signal backward layer by layer, giving each weight its share of the blame.
How it actually works
Backpropagation applies the chain rule from calculus to compute the gradient of the loss with respect to every weight efficiently, in a single backward pass through the network. It’s what makes training deep networks tractable — without it, computing all those derivatives would be hopeless. Paired with an optimizer, it’s the mechanism that turns “the model was wrong” into concrete weight updates.
For product teams
The math that makes learning possible at all; you rarely touch it, but every trained model depends on it.
For engineers
Chain-rule computation of loss gradients w.r.t. all params in one backward pass; feeds the optimizer.
Related
- Gradient — Computes the gradient.
- Gradient descent — Used by gradient descent.
- Loss function — To reduce the loss function.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome