Adversarial Training
Adversarial training aims to enhance the robustness of machine learning models, particularly deep neural networks, against adversarial attacks—malicious inputs designed to cause misclassification. Current research focuses on improving the efficiency and effectiveness of adversarial training methods, exploring techniques like vector quantization for input transformation, null-space projection for gradient optimization, and module-wise adaptive training for end-to-end systems, as well as applying these techniques to various model architectures including LLMs and Vision Transformers. This field is crucial for ensuring the reliability and security of AI systems in real-world applications, particularly in safety-critical domains where model robustness is paramount.
Papers
Power side-channel leakage localization through adversarial training of deep neural networks
Jimmy Gammell, Anand Raghunathan, Kaushik Roy
On the Robustness of Adversarial Training Against Uncertainty Attacks
Emanuele Ledda, Giovanni Scodeller, Daniele Angioni, Giorgio Piras, Antonio Emanuele Cinà, Giorgio Fumera, Battista Biggio, Fabio Roli
Conflict-Aware Adversarial Training
Zhiyu Xue, Haohan Wang, Yao Qin, Ramtin Pedarsani
On the Geometry of Regularization in Adversarial Training: High-Dimensional Asymptotics and Generalization Bounds
Matteo Vilucchio, Nikolaos Tsilivis, Bruno Loureiro, Julia Kempe
LiOn-XA: Unsupervised Domain Adaptation via LiDAR-Only Cross-Modal Adversarial Training
Thomas Kreutz, Jens Lemke, Max Mühlhäuser, Alejandro Sanchez Guinea