cs.LGNov 10, 2026

Structured Confidence Perturbation via Fisher Geometry for Adversarial Training

Authors: Chenlong Wei, Hanling Wang, Chong Zhang, Qinkai Yu, Fangyu Wu, Xiaobo Jin

Organizations: Xi’an Jiaotong-Liverpool University, China · University of California, San Diego, United States · University of Exeter, England

Abstract

Adversarial training is often plagued by "pseudo-attacks", perturbations that significantly increase the loss value but fail to truly alter the model's confidence structure. To address this, we propose two complementary approaches to directly attack and defend against confidence structure deformation. First, building upon the traditional loss maximization perspective, we introduce a hybrid objective that integrates cross-entropy and KL divergence, encouraging adversarial examples to simultaneously disrupt prediction accuracy and distribution consistency. Second, we geometrically reconstruct adversarial training by generating perturbations that maximize the distribution distance on the confidence manifold. Theoretical analysis shows that the optimal perturbation direction aligns with the principal eigenvectors of the Fisher information matrix. We also propose an efficient algorithm to avoid high-dimensional matrix computations. Experiments on multiple datasets and model architectures demonstrate that our method outperforms existing methods in adversarial robustness and robust transferability, particularly excelling in structural perturbations within high-confidence regions.

Figures & tables

Explore similar work

CardsList
  1. Probabilistic Adversarial Training

    Sep 30, 2026Andi Zhang, Xingyu Zhao, Siddartha KhastgirAdversarial TrainingAdversarial Robustness

  2. Information-Geometric Inverse Distillation for Enhancing Adversarial Transferability

    Feb 24, 2025Wenyuan Wu, Yuan Sun, Yingke Chen +4Transfer-Based Adversarial AttacksKnowledge Distillation

  3. Mitigating Error Amplification in Fast Adversarial Training

    Apr 27, 2026Mengnan Zhao, Lihe Zhang, Bo Wang +3Adversarial TrainingNeural Network Generalization