Cyclical Learning Rates for Training Neural Networks
This provides a practical tool for everyone who trains neural networks by simplifying hyperparameter tuning, though it is incremental as it builds on existing learning rate methods.
The paper tackles the problem of tuning learning rates for training deep neural networks by proposing cyclical learning rates, which vary cyclically between bounds instead of decreasing monotonically, resulting in improved classification accuracy without extensive tuning and often in fewer iterations, as demonstrated on datasets like CIFAR-10, CIFAR-100, and ImageNet.
It is known that the learning rate is the most important hyper-parameter to tune for training deep neural networks. This paper describes a new method for setting the learning rate, named cyclical learning rates, which practically eliminates the need to experimentally find the best values and schedule for the global learning rates. Instead of monotonically decreasing the learning rate, this method lets the learning rate cyclically vary between reasonable boundary values. Training with cyclical learning rates instead of fixed values achieves improved classification accuracy without a need to tune and often in fewer iterations. This paper also describes a simple way to estimate "reasonable bounds" -- linearly increasing the learning rate of the network for a few epochs. In addition, cyclical learning rates are demonstrated on the CIFAR-10 and CIFAR-100 datasets with ResNets, Stochastic Depth networks, and DenseNets, and the ImageNet dataset with the AlexNet and GoogLeNet architectures. These are practical tools for everyone who trains neural networks.