Deep Learning without Shortcuts: Shaping the Kernel with Tailored Rectifiers

TL;DR

This paper introduces a novel transformation compatible with Leaky ReLUs that enables training deep vanilla neural networks effectively, achieving competitive accuracy to ResNets without shortcut connections or normalization layers.

Contribution

The authors develop a new transformation method that improves deep vanilla network training with Leaky ReLUs, surpassing previous methods like EOC in accuracy and depth scalability.

Findings

Deep vanilla networks trained with the new method achieve high validation accuracy.

The method is computationally efficient and compatible with Leaky ReLUs.

Validation accuracy remains stable or improves with increasing depth.

Abstract

Training very deep neural networks is still an extremely challenging task. The common solution is to use shortcut connections and normalization layers, which are both crucial ingredients in the popular ResNet architecture. However, there is strong evidence to suggest that ResNets behave more like ensembles of shallower networks than truly deep ones. Recently, it was shown that deep vanilla networks (i.e. networks without normalization layers or shortcut connections) can be trained as fast as ResNets by applying certain transformations to their activation functions. However, this method (called Deep Kernel Shaping) isn't fully compatible with ReLUs, and produces networks that overfit significantly more than ResNets on ImageNet. In this work, we rectify this situation by developing a new type of transformation that is fully compatible with a variant of ReLUs -- Leaky ReLUs. We show in experiments that our method, which introduces negligible extra computational cost, achieves validation accuracies with deep vanilla networks that are competitive with ResNets (of the same width/depth), and significantly higher than those obtained with the Edge of Chaos (EOC) method. And unlike with EOC, the validation accuracies we obtain do not get worse with depth.