AI Foundations
Learning rate
A setting that controls how large each model update is during training.
Example
Engineers lower the learning rate after training begins to jump past good solutions.
Why it matters
It helps balance making useful progress with avoiding disruptive changes during learning.
You'll hear it
“The learning changes are too large; let's try smaller steps.”
Can you control it?
Set by the team
Model builders set or schedule the learning rate during training. Users of a finished model cannot adjust it.
Common questions
- Can the setting change during training?
- Yes. Builders may reduce or otherwise adjust it as learning progresses.
- What happens if the learning rate is too high?
- Training may become unstable or jump past settings that would produce lower error.
- What happens if the learning rate is too low?
- Training can take much longer and may make too little progress.