How to decay your learning rate
arXiv:2103.12682
Abstract
Complex learning rate schedules have become an integral part of deep learning. We find empirically that common fine-tuned schedules decay the learning rate after the weight norm bounces. This leads to the proposal of ABEL: an automatic scheduler which decays the learning rate by keeping track of the weight norm. ABEL's performance matches that of tuned schedules and is more robust with respect to its parameters. Through extensive experiments in vision, NLP, and RL, we show that if the weight norm does not bounce, we can simplify schedules even further with no loss in performance. In such cases, a complex schedule has similar performance to a constant learning rate with a decay at the end of training.
9 + 14 pages, 5 + 11 figures
References in corpus (13)
- Language Models are Few-Shot Learners
- Scaling Laws for Neural Language Models
- Gradient-based Hyperparameter Optimization through Reversible Learning
- L2 Regularization versus Batch and Weight Normalization
- Measuring the Effects of Data Parallelism on Neural Network Training
- Towards Explaining the Regularization Effect of Initial Large Learning Rate in Training Neural Networks
- Learned Optimizers that Scale and Generalize
- Gradient Descent Maximizes the Margin of Homogeneous Neural Networks
- The large learning rate phase of deep learning: the catapult mechanism
- An Exponential Learning Rate Schedule for Deep Learning
- Learning Rate Annealing Can Provably Help Generalization, Even for Convex Problems
- Disentangling Adaptive Gradient Methods from Learning Rates
- Spherical Motion Dynamics: Learning Dynamics of Neural Network with Normalization, Weight Decay, and SGD