Research map / Adversarial robustness and security / Certified and empirical robustness
The Adversarial Attack and Detection under the Fisher Information Metric
Chenxiao Zhao, P. Thomas Fletcher, Mixue Yu, Yaxin Peng, Guixu Zhang, Chaomin Shen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace the input gradient direction with the top Fisher information eigenvector | One-step attack on LeNet-5 · MNIST | mean least ℓ2 perturbation norm: 2.11 -> 1.80 | Improved |
| Replace the input gradient direction with the top Fisher information eigenvector | One-step attack on VGG · CIFAR-10 | fooling rate (%): 95.21 -> 97.85 | Improved |
| Use Fisher information eigenvalues as detector features instead of combined kernel density and Bayesian uncertainty features | Random-forest adversarial detector · CIFAR-10, Opt attacks | AUC (%): 93.77 -> 99.45 | Improved |
| Use Fisher information eigenvalues as detector features instead of combined kernel density and Bayesian uncertainty features | Random-forest adversarial detector · CIFAR-10, OTCM attacks | AUC (%): 94.45 -> 93.68 | Hurt |
| Use Fisher information eigenvalues as detector features instead of combined kernel density and Bayesian uncertainty features | Random-forest adversarial detector · CIFAR-10, OSSA attacks | AUC (%): 93.54 -> 98.01 | Improved |
| Train the detector on OSSA rather than other attack types | Random-forest detector · MNIST cross-attack generalization | AUC (%): 89.47 | Hurt |
| Use alias sampling with power iteration instead of exact eigen-decomposition | OSSA computation · CIFAR-10 | time (seconds): 1.49±{0.12} -> 0.25±{0.01} | Improved |
Closest papers by meaning
- Adversarial Robustness Via Fisher-Rao Regularization (TPAMI 2022 · similarity 0.87)
- On the Existence of Consistent Adversarial Attacks in High-Dimensional Linear Classification (ICML 2026 · similarity 0.85)
- Fisher-Rao Sensitivity for Out-of-Distribution Detection in Deep Neural Networks (ICLR 2026 · similarity 0.85)
- Adversarial Fisher Vectors for Unsupervised Representation Learning (NeurIPS 2019 · similarity 0.84)
- Metric Learning for Adversarial Robustness (NeurIPS 2019 · similarity 0.84)
- Detecting Adversarial Data Using Perturbation Forgery (CVPR 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
