Research map / Adversarial robustness and security / Black-box and transfer attacks
Robust Detection of Adversarial Attacks by Modeling the Intrinsic Properties of Deep Neural Networks
Zhihao Zheng, Pengyu Hong
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 |
|---|---|---|---|
| model hidden-state distributions with GMM (I-defender) vs Defense-GAN with 10 runs x 50 iterations | detector · FGSM MNIST eps=0.3 | detection AUC: 0.993 vs 0.985 | Mixed |
| I-defender vs Defense-GAN at small perturbation | detector · FGSM F-MNIST eps=0.1 | detection AUC: 0.775 -> 0.9302 | Improved |
| I-defender vs Defense-GAN at large perturbation | detector · FGSM MNIST eps=0.3 | detection AUC: 0.999 vs 0.993 | No clear effect |
| unsupervised I-defender vs supervised Subnetwork on unseen attack (Iter-l2) | detector · CIFAR-10 semi white-box | detection AUC: 68.980 -> 87.464 | Improved |
| I-defender vs supervised Subnetwork on the training attack Iter-linf | detector · CIFAR-10 semi white-box | detection AUC: 98.660 vs 79.398 | Hurt |
| attacker has defense parameters (source DNN) vs unknown parameters (target DNN) | I-defender · CIFAR-10 gray-box Iter-linf eps=0.013 | detection AUC: 83.261 -> 58.967 | Hurt |
| larger classifier WRN46-8 instead of WRN34-8 | I-defender · CIFAR-10 gray-box Iter-linf eps=0.013 | detection AUC: 85.203 -> 87.549 | Improved |
| increase attacker penalty weight alpha from 1e-3 to 1 | I-defender · CIFAR-10 gray-box target DNN Iter-linf eps=0.013 | detection AUC: 85.203 -> 83.945 | No clear effect |
| FGSM attack on ImageNet | I-defender · ImageNet gray-box FGSM eps=0.3 | detection AUC: 62.685 vs 97.201 (Iter-linf) | Hurt |
Closest papers by meaning
- Towards Deep Learning Models Resistant to Adversarial Attacks (ICLR 2018 · similarity 0.90)
- A Direct Approach to Robust Deep Learning Using Adversarial Networks (ICLR 2019 · similarity 0.89)
- A Self-supervised Approach for Adversarial Robustness (CVPR 2020 · similarity 0.89)
- Increasing Confidence in Adversarial Robustness Evaluations (NeurIPS 2022 · similarity 0.89)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.88)
- Interpreting Robustness Proofs of Deep Neural Networks (ICLR 2024 · similarity 0.87)
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.
