A Comprehensive and Modularized Statistical Framework for Gradient Norm Equality in Deep Neural Networks

TL;DR

This paper introduces a new modular statistical framework and a novel metric called Block Dynamical Isometry to evaluate gradient norm behavior in deep neural networks, enabling better analysis and improvements of network initialization, normalization, and architecture.

Contribution

It proposes a weak-assumption, norm-based metric and a modular framework for analyzing gradient norms in complex DNNs, along with practical normalization and initialization strategies.

Findings

Gradient Norm Equality is a universal principle in DNNs.

Second moment normalization is 30% faster than batch normalization.

The framework effectively analyzes diverse network components and structures.

Abstract

In recent years, plenty of metrics have been proposed to identify networks that are free of gradient explosion and vanishing. However, due to the diversity of network components and complex serial-parallel hybrid connections in modern DNNs, the evaluation of existing metrics usually requires strong assumptions, complex statistical analysis, or has limited application fields, which constraints their spread in the community. In this paper, inspired by the Gradient Norm Equality and dynamical isometry, we first propose a novel metric called Block Dynamical Isometry, which measures the change of gradient norm in individual block. Because our Block Dynamical Isometry is norm-based, its evaluation needs weaker assumptions compared with the original dynamical isometry. To mitigate the challenging derivation, we propose a highly modularized statistical framework based on free probability. Our framework includes several key theorems to handle complex serial-parallel hybrid connections and a library to cover the diversity of network components. Besides, several sufficient prerequisites are provided. Powered by our metric and framework, we analyze extensive initialization, normalization, and network structures. We find that Gradient Norm Equality is a universal philosophy behind them. Then, we improve some existing methods based on our analysis, including an activation function selection strategy for initialization techniques, a new configuration for weight normalization, and a depth-aware way to derive coefficients in SeLU. Moreover, we propose a novel normalization technique named second moment normalization, which is theoretically 30% faster than batch normalization without accuracy loss. Last but not least, our conclusions and methods are evidenced by extensive experiments on multiple models over CIFAR10 and ImageNet.