Learning an Adaptive Learning Rate Schedule

TL;DR

This paper introduces a reinforcement learning framework that automatically learns an adaptive learning rate schedule, improving training efficiency and generalization across different neural network architectures and datasets.

Contribution

It presents a novel reinforcement learning based method for automatically adapting learning rates during training, surpassing traditional fixed or predefined schedules.

Findings

Auto-learned schedule improves test accuracy

Controller generalizes across datasets and architectures

Outperforms fixed and predefined schedules

Abstract

The learning rate is one of the most important hyper-parameters for model training and generalization. However, current hand-designed parametric learning rate schedules offer limited flexibility and the predefined schedule may not match the training dynamics of high dimensional and non-convex optimization problems. In this paper, we propose a reinforcement learning based framework that can automatically learn an adaptive learning rate schedule by leveraging the information from past training histories. The learning rate dynamically changes based on the current training dynamics. To validate this framework, we conduct experiments with different neural network architectures on the Fashion MINIST and CIFAR10 datasets. Experimental results show that the auto-learned learning rate controller can achieve better test results. In addition, the trained controller network is generalizable -- able to be trained on one data set and transferred to new problems.