AI Blog

AI Blog

by Michele Laurelli

Back to Glossary

Learning Rate

Concept
Definition

Hyperparameter controlling how much to adjust model weights during training.

Step size in gradient descent. Too high: unstable training. Too low: slow convergence. Typical values: 0.001, 0.01, 0.1.

Examples

1

Learning rate 0.001

2

LR too high causes divergence

3

Adaptive learning rates

Related Terms

Gradient Descent

An optimization algorithm that iteratively adjusts parameters to minimize a loss function by following the gradient.

Hyperparameter

A configuration variable that is set before training and controls the learning process.

Michele Laurelli - AI Research & Engineering