L4: Practical loss-based stepsize adaptation for deep learning

Abstract

We propose a stepsize adaptation scheme for stochastic gradient descent. It operates directly with the loss function and rescales the gradient in order to make fixed predicted progress on the loss. We demonstrate its capabilities by conclusively improving the performance of Adam and Momentum optimizers. The enhanced optimizers with default hyperparameters consistently outperform their constant stepsize counterparts, even the best ones, without a measurable increase in computational cost. The performance is validated on multiple architectures including dense nets, CNNs, ResNets, and the recurrent Differential Neural Computer on classical datasets MNIST, fashion MNIST, CIFAR10 and others.

arXiv commentNeurIPS, 2018

Cited by in corpus (8)

• Stochastic Polyak Step-size for SGD: An Adaptive Learning Rate for Fast Convergence
• Painless Stochastic Gradient: Interpolation, Line-Search, and Convergence Rates
• SGD for Structured Nonconvex Functions: Learning Rates, Minibatching and Interpolation
• Training Neural Networks for and by Interpolation
• LOSSGRAD: automatic learning rate in gradient descent
• Collaborative Sampling in Generative Adversarial Networks
• Step Size Matters in Deep Learning
• Parabolic Approximation Line Search for DNNs