Defensive Distillation

Defensive distillation is a technique used to enhance the robustness of deep neural networks (DNNs) against adversarial attacks, which involve subtly altering inputs to cause misclassification. Current research focuses on improving defensive distillation's effectiveness against increasingly sophisticated attacks, exploring its application with various architectures like Vision Transformers (ViTs) and Convolutional Neural Networks (CNNs), and integrating it with other defense mechanisms such as adversarial training and denoising autoencoders. This work is significant because it addresses the critical vulnerability of DNNs to malicious manipulation, impacting the reliability and security of AI systems in diverse applications, from medical image analysis to wireless network security.

Papers

October 21, 2024

Model Mimic Attack: Knowledge Distillation for Provably Transferable Adversarial Examples
Kirill Lukyanov, Andrew Perminov, Denis Turdakov, Mikhail Pautov
Adversarial Attack Knowledge Distillation Adversarial Example Adversarial Perturbation Model Attack Defensive Distillation

April 5, 2024

Evaluating Adversarial Robustness: A Comparison Of FGSM, Carlini-Wagner Attacks, And The Role of Distillation as Defense Mechanism
Trilokesh Ranjan Sarkar, Nilanjan Das, Pralay Sankar Maitra, Bijoy Some, Ritwik Saha, Orijita Adhikary, Bishal Bose, Jaydip Sen
Deep Neural Network Adversarial Attack Adversarial Robustness Mutual Distillation DEfense Mechanism Fast Gradient Sign Method Based Attack Defensive Distillation

August 15, 2023

SEDA: Self-Ensembling ViT with Defensive Distillation and Adversarial Training for robust Chest X-rays Classification
Raza Imam, Ibrahim Almakky, Salma Alrashdi, Baketah Alrashdi, Mohammad Yaqub
Vision Transformer Adversarial Example Adversarial Training Adversarial Perturbation Chest X Ray Adversarial Distillation Defensive Distillation

May 14, 2023

Improving Defensive Distillation using Teacher Assistant
Maniratnam Mandal, Suna Gao
Adversarial Attack Knowledge Distillation Adversarial Sample Distilled Model Teaching Assistant Defensive Distillation

March 28, 2023

Denoising Autoencoder-based Defensive Distillation as an Adversarial Robustness Algorithm
Bakary Badjie, José Cecílio, António Casimiro
Adversarial Attack Adversarial Robustness Poisoning Attack 3D Object Detection Distillation Defensive Distillation

August 12, 2022

April 1, 2022

Preventing Distillation-based Attacks on Neural Network IP
Mahdieh Grailoo, Zain Ul Abideen, Mairo Leier, Samuel Pagliarini
Obfuscation Technique Defensive Distillation

February 16, 2022

The Adversarial Security Mitigations of mmWave Beamforming Prediction Models using Defensive Distillation and Adversarial Retraining
Murat Kuzlu, Ferhat Ozgur Catak, Umit Cali, Evren Catak, Ozgur Guler
Deep Neural Network Adversarial Attack Adversarial Example mmWave Signal Prediction Model Beam Prediction Adversarial Retraining Defensive Distillation

November 3, 2021

LTD: Low Temperature Distillation for Robust Adversarial Training
Erh-Chung Chen, Che-Rung Lee
Adversarial Attack Adversarial Training Low Temperature Distillation Gradient Masking Defensive Distillation Robust Adversarial Training

Defensive Distillation

Papers

Model Mimic Attack: Knowledge Distillation for Provably Transferable Adversarial Examples

Evaluating Adversarial Robustness: A Comparison Of FGSM, Carlini-Wagner Attacks, And The Role of Distillation as Defense Mechanism

SEDA: Self-Ensembling ViT with Defensive Distillation and Adversarial Training for robust Chest X-rays Classification

Improving Defensive Distillation using Teacher Assistant

Denoising Autoencoder-based Defensive Distillation as an Adversarial Robustness Algorithm

A Knowledge Distillation-Based Backdoor Attack in Federated Learning

Defensive Distillation based Adversarial Attacks Mitigation Method for Channel Estimation using Deep Learning Models in Next-Generation Wireless Networks

Preventing Distillation-based Attacks on Neural Network IP

The Adversarial Security Mitigations of mmWave Beamforming Prediction Models using Defensive Distillation and Adversarial Retraining

LTD: Low Temperature Distillation for Robust Adversarial Training