Glossary · Phase 4 · Deep learning
Closure
A small function that remembers the values around it when it was made. Autograd stores one per operation to run its backward step later.
Try it now
This is all of backpropagation: one small rule per operation, and the right order.
Learn it properly
Build an autograd engine · Write backprop yourself in ~100 lines (micrograd). About 40 minutes.
Other terms from the same lesson
- autograd: Software that works out gradients automatically from the computation graph, so you never do it by hand.
Never trained an AI?
Teach a small one to write, right in your browser. It takes about 20 seconds and needs no account.