Stochastic Defense

Stochastic defense techniques aim to improve the robustness of machine learning models, particularly against adversarial attacks that manipulate inputs to cause misclassification or incorrect predictions. Current research focuses on evaluating the effectiveness of various stochastic methods across different model architectures, including vision transformers and large language models, and exploring techniques like adversarial training, input preprocessing, and Monte Carlo simulations to enhance model resilience. Understanding the limitations of these defenses, such as the trade-off between robustness and model performance, and identifying when gradient obfuscation provides a false sense of security, are crucial areas of ongoing investigation with implications for the security and reliability of AI systems in real-world applications.

Papers

October 31, 2024

Noise as a Double-Edged Sword: Reinforcement Learning Exploits Randomized Defenses in Neural Networks
Steve Bakos, Pooria Madani, Heidar Davoudi
Neural Network Adversarial Example Industrial Disturbing Noise Stochastic Defense Perturbation Defense

September 12, 2023

Exploring Non-additive Randomness on ViT against Query-Based Black-Box Attacks
Jindong Gu, Fangyun Wei, Philip Torr, Han Hu
Deep Neural Network Vision Transformer Native Robustness Black Box Attack ViT Lens Stochastic Defense Non Additive

September 1, 2023

Baseline Defenses for Adversarial Attacks Against Aligned Language Models
Neel Jain, Avi Schwarzschild, Yuxin Wen, Gowthami Somepalli, John Kirchenbauer, Ping-yeh Chiang, Micah Goldblum, Aniruddha Saha, Jonas Geiping, Tom Goldstein
Language Model Adversarial Attack Adversarial Training Adversarial Learning Stochastic Defense

May 13, 2023

On the Computational Cost of Stochastic Security
Noah A. Crum, Leanto Sunny, Pooya Ronagh, Raymond Laflamme, Radhakrishnan Balu, George Siopsis
Langevin Dynamic Markov Chain Monte Carlo Energy Based Model Computational Cost Gibbs Algorithm Monte Carlo Simulation Stochastic Defense

February 20, 2023

Efficient Algorithms for Boundary Defense with Heterogeneous Defenders
Si Wei Feng, Jingjin Yu
Efficient Algorithm Cyber Attack Attack Success Rate Multiple Defender Multiple Attack Heterogeneous Zero Day Attack Stochastic Defense

August 29, 2022

Reducing Certified Regression to Certified Classification for General Poisoning Attacks
Zayd Hammoudeh, Daniel Lowd
Native Robustness Novel Regression Poisoning Attack Model Robustness Adversarial Instance Stochastic Defense

June 19, 2022

On the Limitations of Stochastic Pre-processing Defenses
Yue Gao, Ilia Shumailov, Kassem Fawaz, Nicolas Papernot
Adversarial Example Adversarial Robustness Fundamental Limitation Stochastic Way Adversarial Input Stochastic Defense

June 3, 2022

Gradient Obfuscation Checklist Test Gives a False Sense of Security
Nikola Popovic, Danda Pani Paudel, Thomas Probst, Luc Van Gool
Adversarial Attack Security Related False Sense Gradient Obfuscation Stochastic Defense