Intelligent gradient amplification for deep neural networks

Abstract

Deep learning models offer superior performance compared to other machine learning techniques for a variety of tasks and domains, but pose their own challenges. In particular, deep learning models require larger training times as the depth of a model increases, and suffer from vanishing gradients. Several solutions address these problems independently, but there have been minimal efforts to identify an integrated solution that improves the performance of a model by addressing vanishing gradients, as well as accelerates the training process to achieve higher performance at larger learning rates. In this work, we intelligently determine which layers of a deep learning model to apply gradient amplification to, using a formulated approach that analyzes gradient fluctuations of layers during training. Detailed experiments are performed for simpler and deeper neural networks using two different intelligent measures and two different thresholds that determine the amplification layers, and a training strategy where gradients are amplified only during certain epochs. Results show that our amplification offers better performance compared to the original models, and achieves accuracy improvement of around 2.5% on CIFAR- 10 and around 4.5% on CIFAR-100 datasets, even when the models are trained with higher learning rates.