Adversarial Parameter

Adversarial parameter attacks explore vulnerabilities in machine learning models by subtly altering their internal parameters, rather than manipulating input data. Research focuses on developing methods to find these adversarial parameters, often framing the problem as a game between an attacker seeking to reduce model robustness and a defender aiming to maintain accuracy. This line of research is crucial for assessing and improving the security and reliability of deep neural networks, particularly in safety-critical applications where model robustness is paramount. Current work investigates efficient algorithms for finding these parameters and analyzing the conditions under which such attacks are most effective.

Papers

August 14, 2024

Enhancing Adversarial Attacks via Parameter Adaptive Adversarial Attack
Zhibo Jin, Jiayu Zhang, Zhiyu Zhu, Chenyu Zhang, Jiahao Huang, Jianlong Zhou, Fang Chen
Adversarial Attack Adversarial Sample Adversarial Parameter

July 19, 2022

Feasible Adversarial Robust Reinforcement Learning for Underspecified Environments
JB Lanier, Stephen McAleer, Pierre Baldi, Roy Fox
Robust Policy Robust Reinforcement Learning Adversarial Reinforcement Learning Adversarial Parameter

March 20, 2022

Adversarial Parameter Attack on Deep Neural Networks
Lijia Yu, Yihan Wang, Xiao-Shan Gao
Deep Neural Network Perturbation Attack Adversarial Parameter

Adversarial Parameter

Papers

Enhancing Adversarial Attacks via Parameter Adaptive Adversarial Attack

Feasible Adversarial Robust Reinforcement Learning for Underspecified Environments

Adversarial Parameter Attack on Deep Neural Networks