An Adaptive and Momental Bound Method for Stochastic Learning

TL;DR

This paper introduces AdaMod, an adaptive learning rate method that stabilizes training by bounding learning rates based on exponential moving averages, improving convergence especially on complex neural networks.

Contribution

The paper proposes AdaMod, a novel adaptive learning rate method that prevents excessively large updates, enhancing stability and performance in deep neural network training.

Findings

AdaMod eliminates extremely large learning rates during training.

AdaMod significantly improves training stability on DenseNet and Transformer models.

Experimental results show AdaMod outperforms Adam in complex network training.

Abstract

Training deep neural networks requires intricate initialization and careful selection of learning rates. The emergence of stochastic gradient optimization methods that use adaptive learning rates based on squared past gradients, e.g., AdaGrad, AdaDelta, and Adam, eases the job slightly. However, such methods have also been proven problematic in recent studies with their own pitfalls including non-convergence issues and so on. Alternative variants have been proposed for enhancement, such as AMSGrad, AdaShift and AdaBound. In this work, we identify a new problem of adaptive learning rate methods that exhibits at the beginning of learning where Adam produces extremely large learning rates that inhibit the start of learning. We propose the Adaptive and Momental Bound (AdaMod) method to restrict the adaptive learning rates with adaptive and momental upper bounds. The dynamic learning rate bounds are based on the exponential moving averages of the adaptive learning rates themselves, which smooth out unexpected large learning rates and stabilize the training of deep neural networks. Our experiments verify that AdaMod eliminates the extremely large learning rates throughout the training and brings significant improvements especially on complex networks such as DenseNet and Transformer, compared to Adam. Our implementation is available at: https://github.com/lancopku/AdaMod