A technique that reduces the learning rate over time during training to help the model converge more smoothly and avoid overshooting minima.
Last updated 10 months ago