Why Warmup the Learning Rate? Underlying Mechanisms and Improvements

It is common in deep learning to warm up the learning rate η, often by a linear schedule between ηᵢₙᵢₜ = 0 and a predetermined target ηtrgt. In this paper, we show through systematic experiments using SGD and Adam that the overwhelming benefit of warmup arises from allowing the network to tolerate larger ηtrgt by forcing the network to more well-conditioned areas of the loss landscape. The ability to handle larger ηtrgt makes hyperparameter tuning more robust while improving the final performance. We uncover different regimes of operation during the warmup period, depending on whether training starts off in a progressive sharpening or sharpness reduction phase, which in turn depends on the initialization and parameterization. Using these insights, we show how ηᵢₙᵢₜ can be properly chosen by utilizing the loss catapult mechanism, which saves on the number of warmup steps, in some cases completely eliminating the need for warmup. We also suggest an initialization for the variance in Adam which provides benefits similar to warmup.