Regularization in ResNet with Stochastic Depth

TL;DR

This paper analyzes the regularization effects of Stochastic Depth in ResNets using a hybrid theoretical approach, providing guidelines for selecting survival rates to improve model generalization.

Contribution

It offers a novel hybrid analysis combining perturbation and signal propagation to understand SD's regularization effects and guides optimal survival rate choices.

Findings

Provides theoretical insights into SD's regularization mechanisms

Derives guidelines for choosing survival rates in SD training

Enhances understanding of SD's impact on ResNet generalization

Abstract

Regularization plays a major role in modern deep learning. From classic techniques such as L1,L2 penalties to other noise-based methods such as Dropout, regularization often yields better generalization properties by avoiding overfitting. Recently, Stochastic Depth (SD) has emerged as an alternative regularization technique for residual neural networks (ResNets) and has proven to boost the performance of ResNet on many tasks [Huang et al., 2016]. Despite the recent success of SD, little is known about this technique from a theoretical perspective. This paper provides a hybrid analysis combining perturbation analysis and signal propagation to shed light on different regularization effects of SD. Our analysis allows us to derive principled guidelines for choosing the survival rates used for training with SD.