English

Don't Decay the Learning Rate, Increase the Batch Size

Machine Learning 2018-02-27 v2 Computer Vision and Pattern Recognition Distributed, Parallel, and Cluster Computing Machine Learning

Abstract

It is common practice to decay the learning rate. Here we show one can usually obtain the same learning curve on both training and test sets by instead increasing the batch size during training. This procedure is successful for stochastic gradient descent (SGD), SGD with momentum, Nesterov momentum, and Adam. It reaches equivalent test accuracies after the same number of training epochs, but with fewer parameter updates, leading to greater parallelism and shorter training times. We can further reduce the number of parameter updates by increasing the learning rate ϵ\epsilon and scaling the batch size BϵB \propto \epsilon. Finally, one can increase the momentum coefficient mm and scale B1/(1m)B \propto 1/(1-m), although this tends to slightly reduce the test accuracy. Crucially, our techniques allow us to repurpose existing training schedules for large batch training with no hyper-parameter tuning. We train ResNet-50 on ImageNet to 76.1%76.1\% validation accuracy in under 30 minutes.

Keywords

Cite

@article{arxiv.1711.00489,
  title  = {Don't Decay the Learning Rate, Increase the Batch Size},
  author = {Samuel L. Smith and Pieter-Jan Kindermans and Chris Ying and Quoc V. Le},
  journal= {arXiv preprint arXiv:1711.00489},
  year   = {2018}
}

Comments

11 pages, 8 figures. Published as a conference paper at ICLR 2018

R2 v1 2026-06-22T22:33:23.961Z