Reference · Glossary

Gradient descent

An optimization loop that **nudges model weights** in the direction that reduces loss — the core “learning” step in many trainers.

When to use

Explaining how neural nets and linear models improve iteratively on a loss function.

When not to

As a hand-wavy synonym for “AI magic” — tie it to a concrete loss and learning rate.

Example

Compute loss on a batch → backprop gradients → `weight -= lr * gradient` → repeat until loss plateaus.