Are L2 adversarial examples intrinsically different?

2020/02/28 by Mingxuan Li, Jingyuan Wang, Li, Mingxuan +3
Computer Science · Engineering · #Adversarial Robustness in Machine Learning #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Integrated Circuits and Semiconductor Failure Analysis #Ion-surface interactions and analysis

paper · pdf · doi:10.48550/arxiv.2002.12527

openalex publication_date 2020/02/28 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Deep Neural Network (DDN) has achieved notable success in various tasks, including many security concerning scenarios. However, a considerable amount of work has proved its vulnerability to adversaries. We unravel the properties that can intrinsically differentiate adversarial examples and normal inputs through theoretical analysis. That is, adversarial examples generated by L₂ attacks usually have larger input sensitivity which can be used to identify them efficiently. We also found that those generated by L_∞ attacks will be different enough in the pixel domain to be detected empirically. To verify our analysis, we proposed a Guided Complementary Defense module (GCD) integrating detection and recovery processes. When compared with adversarial detection methods, our detector achieves a detection AUC of over 0.98 against most of the attacks. When comparing our guided rectifier with commonly used adversarial training methods and other rectification methods, our rectifier outperforms them by a large margin. We achieve a recovered classification accuracy of up to 99% on MNIST, 89% on CIFAR-10, and 87% on ImageNet subsets against L₂ attacks. Furthermore, under the white-box setting, our holistic defensive module shows a promising degree of robustness. Thus, we confirm that at least L₂ adversarial examples are intrinsically different enough from normal inputs both theoretically and empirically. And we shed light upon designing simple yet effective defensive methods with these properties.

Citations

• Intriguing properties of neural networks
• Gradient-based learning applied to document recognition
• Transferability in Machine Learning: from Phenomena to Black-Box Attacks using Adversarial Samples
• On Detecting Adversarial Perturbations
• Gradient Similarity: An Explainable Approach to Detect Adversarial\n Attacks against Deep Learning
• PixelDefend: Leveraging Generative Models to Understand and Defend against Adversarial Examples
• Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examples
• Defense-GAN: Protecting Classifiers Against Adversarial Attacks Using Generative Models
• A New Defense Against Adversarial Images: Turning a Weakness into a Strength
• Learning Multiple Layers of Features from Tiny Images

Related

• Not So Robust After All: Evaluating the Robustness of Deep Neural Networks to Unseen Adversarial Attacks
• Generating Unrestricted Adversarial Examples via Three Parameters
• Exploiting the Inherent Limitation of L0 Adversarial Examples
• A Few Large Shifts: Layer-Inconsistency Based Minimal Overhead Adversarial Example Detection
• A Learning and Masking Approach to Secure Learning