Scheduler
Last Updated: July 29, 2026 | By Mihail Sebastian | AI Dictionary
A rule that changes the learning rate during training, lowering it over time so a model takes big steps early and fine adjustments near the end.
What is a Scheduler?
In machine learning, a scheduler is a rule that changes the learning rate over the course of training, usually lowering it so the model takes big steps early and fine adjustments late.
The same word names other things in computing, such as job schedulers that launch tasks at set times and operating-system schedulers that share CPU between processes. In an ML context it almost always means the learning-rate kind.
How a Scheduler Works
A single fixed learning rate forces a compromise. High enough to make early progress, it is too coarse to settle into a good minimum at the end; low enough for the end, it makes early training crawl.
A schedule removes the compromise by giving each phase of training the rate it needs. The schedule’s shape and timing are themselves hyperparameters, chosen before training begins, and the optimizer applies whatever rate the schedule currently dictates.
Types of Learning Rate Schedulers
- Step decay cuts the learning rate by a fixed factor at chosen epochs, for example dividing by 10 after epoch 30 and again after epoch 60.
- Cosine annealing slides the rate down smoothly along a cosine curve from its starting value toward zero.
- Reduce-on-plateau watches a validation metric and lowers the rate only when improvement stalls.
- Warmup starts the rate near zero and ramps it up over the first steps before handing off to a decay schedule. Transformer training made this combination standard.
Example of a Scheduler
A standard recipe for training an image classifier uses step decay. Training starts at a learning rate of 0.1, then divides it by 10 at epoch 30 and again at epoch 60.
The learning curve shows the schedule at work. Accuracy climbs, plateaus, then jumps visibly at each drop, because the smaller steps let the model settle into fine detail of the loss surface it could previously only skip across.
Related AI terms: Learning Rate · Hyperparameter · Optimization · Optimizer · Epoch
Did you like the Scheduler gist?
Learn about 250+ need-to-know artificial intelligence terms in the AI Dictionary.
Mihail Sebastian — Writes about AI governance, regulation, and the technology behind them. Placeholder bio — replace with a real credential line. About