Defensive Distillation is Not Robust to Adversarial Examples

Abstract

We show that defensive distillation is not secure: it is no more resistant to targeted misclassification attacks than unprotected neural networks.

Cited by in corpus (44)

• On Evaluating Adversarial Robustness
• Fast is better than free: Revisiting adversarial training
• NO Need to Worry about Adversarial Examples in Object Detection in Autonomous Vehicles
• Adversarial Examples that Fool Detectors
• TAD: Transfer Learning-based Multi-Adversarial Detection of Evasion Attacks against Network Intrusion Detection Systems
• Adversarial Examples in Modern Machine Learning: A Review
• Standard detectors aren't (currently) fooled by physical adversarial stop signs
• Assessing Threat of Adversarial Examples on Deep Neural Networks
• Wasserstein Smoothing: Certified Robustness against Wasserstein Adversarial Attacks
• RAID: Randomized Adversarial-Input Detection for Neural Networks
• Online Robustness Training for Deep Reinforcement Learning
• Building Robust Deep Neural Networks for Road Sign Detection
• Using Non-invertible Data Transformations to Build Adversarial-Robust Neural Networks
• Image Transformation can make Neural Networks more robust against Adversarial Examples
• ShapeAdv: Generating Shape-Aware Adversarial 3D Point Clouds
• Evaluating the Robustness of Geometry-Aware Instance-Reweighted Adversarial Training
• Regula Sub-rosa: Latent Backdoor Attacks on Deep Neural Networks
• Robustness Certificates Against Adversarial Examples for ReLU Networks
• Trust but Verify: An Information-Theoretic Explanation for the Adversarial Fragility of Machine Learning Systems, and a General Defense against Adversarial Attacks
• Bandlimiting Neural Networks Against Adversarial Attacks
• Minimax Defense against Gradient-based Adversarial Attacks
• FAT: Federated Adversarial Training
• Adequacy of the Gradient-Descent Method for Classifier Evasion Attacks
• Efficient detection of adversarial images
• Multi-objective Search of Robust Neural Architectures against Multiple Types of Adversarial Attacks
• On the human-recognizability phenomenon of adversarially trained deep image classifiers
• Non-Determinism in Neural Networks for Adversarial Robustness
• Controlled Caption Generation for Images Through Adversarial Attacks
• Who is Responsible for Adversarial Defense?
• Ensemble Generative Cleaning with Feedback Loops for Defending Adversarial Attacks
• A Useful Taxonomy for Adversarial Robustness of Neural Networks
• An Information-Theoretic Explanation for the Adversarial Fragility of AI Classifiers
• An Empirical Study of DNNs Robustification Inefficacy in Protecting Visual Recommenders
• Attack to Fool and Explain Deep Networks
• Rethinking Uncertainty in Deep Learning: Whether and How it Improves Robustness
• Delving into the pixels of adversarial samples
• Strategies to architect AI Safety: Defense to guard AI from Adversaries
• A New Family of Neural Networks Provably Resistant to Adversarial Attacks
• Orthogonal Deep Models As Defense Against Black-Box Attacks
• Tricking Adversarial Attacks To Fail
• Enhancing Resilience of Deep Learning Networks by Means of Transferable Adversaries
• Can Intelligent Hyperparameter Selection Improve Resistance to Adversarial Examples?
• Target Training Does Adversarial Training Without Adversarial Samples
• Group-Structured Adversarial Training