Short Definition
The learning rate controls how large each optimization update is during training.
Intuition
A small learning rate learns slowly but steadily. A large learning rate moves faster but can bounce around or diverge.
Technical Definition
The learning rate is the scalar step size applied to gradient-based parameter updates.
Example
If training loss explodes, lowering the learning rate is often one of the first checks.
Common Misunderstandings
There is no universal best learning rate.
Learning-rate schedules can matter as much as the initial value.