On the Robustness of Convolutional Neural Networks to Internal Architecture and Weight Perturbations

TL;DR

This paper investigates how convolutional neural networks respond to internal perturbations, revealing that higher layers are more robust than initial layers, which are highly fragile under weight and architecture disruptions.

Contribution

It provides a systematic analysis of CNN robustness to internal weight and architecture perturbations, highlighting layer-specific vulnerabilities and resilience.

Findings

Higher convolutional layers are robust to weight removal.

First convolutional layer is highly fragile under perturbations.

Performance drops significantly with perturbations in early layers.

Abstract

Deep convolutional neural networks are generally regarded as robust function approximators. So far, this intuition is based on perturbations to external stimuli such as the images to be classified. Here we explore the robustness of convolutional neural networks to perturbations to the internal weights and architecture of the network itself. We show that convolutional networks are surprisingly robust to a number of internal perturbations in the higher convolutional layers but the bottom convolutional layers are much more fragile. For instance, Alexnet shows less than a 30% decrease in classification performance when randomly removing over 70% of weight connections in the top convolutional or dense layers but performance is almost at chance with the same perturbation in the first convolutional layer. Finally, we suggest further investigations which could continue to inform the robustness of convolutional networks to internal perturbations.