On the Interplay of Convolutional Padding and Adversarial Robustness

TL;DR

This paper investigates how different padding modes in CNNs influence adversarial robustness, revealing that boundary perturbations caused by attacks are affected by padding choices, impacting model security.

Contribution

It provides an analysis of the relationship between padding strategies and adversarial vulnerability, highlighting the importance of padding in robustness.

Findings

Padding affects boundary perturbation patterns in adversarial attacks.

Different padding modes influence the robustness of CNNs.

Boundary anomalies are linked to padding choices.

Abstract

It is common practice to apply padding prior to convolution operations to preserve the resolution of feature-maps in Convolutional Neural Networks (CNN). While many alternatives exist, this is often achieved by adding a border of zeros around the inputs. In this work, we show that adversarial attacks often result in perturbation anomalies at the image boundaries, which are the areas where padding is used. Consequently, we aim to provide an analysis of the interplay between padding and adversarial attacks and seek an answer to the question of how different padding modes (or their absence) affect adversarial robustness in various scenarios.