Research map / Adversarial robustness and security / Certified and empirical robustness
Robust Learning against Relational Adversaries
Yizhen Wang, Mohannad Alhanahnah, Xiaozhu Meng, Ke Wang, Mihai Christodorescu, Somesh Jha
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 |
|---|---|---|---|
| add normalization (N&P) to RF | Vanilla-RF · Authorship, non-adaptive attack | accuracy (%): 13.2±3.3 -> 72.3±4.0 | Improved |
| add normalization (N&P) to LSTM | Vanilla-LSTM · Authorship, non-adaptive attack | accuracy (%): 21.1±2.8 -> 73.7±3.6 | Improved |
| add normalization to RF under adaptive attack | Vanilla-RF · Authorship, adaptive attack | accuracy (%): 13.2±3.3 -> 70.5±4.6 | Improved |
| add normalization to LSTM under Adaptive+ attack (7 unnormalized transformations) | Vanilla-LSTM · Authorship, Adaptive+ attack | accuracy (%): 0.9±0.4 -> 49.9±7.4 | Improved |
| add normalization to RF under Adaptive+ attack | Vanilla-RF · Authorship, Adaptive+ attack | accuracy (%): 1.0±0.4 -> 37.3±13.3 | Improved |
| add normalization to RF on clean inputs | Vanilla-RF · Authorship, clean | accuracy (%): 90.4±1.7 -> 76.1±3.8 | Hurt |
| add normalization to LSTM on clean inputs | Vanilla-LSTM · Authorship, clean | accuracy (%): 88.4±3.7 -> 78.7±4.8 | Hurt |
| replace adversarial training with N&P-LSTM | Adv-LSTM · Authorship, Adaptive+ attack | accuracy (%): 25.2±5.1 -> 49.9±7.4 | Improved |
| replace adversarial training with normalization | Adv-LSTM · Authorship, adaptive attack | accuracy (%): 30.8±4.0 -> 71.2±4.8 | Improved |
| replace adversarial training with normalization (training cost) | Adv-LSTM · Authorship training time | days / hours: more than 20 days -> less than 12 hours overhead | Improved |
| add normalization before adversarial training (Unified) vs adversarial training alone | Adv-Trained · Sleipnir adversarial inputs | FNR (%): 27.9±8.2 -> 5.5±0.5 | Improved |
| add API substitution normalization on top of Al-Dujaili adversarial training | Al-Dujaili et al. [2018] · Sleipnir adversarial inputs | FNR (%): 89.9±7.8 -> 5.5±0.5 | Improved |
Closest papers by meaning
- Better Safe Than Sorry: Preventing Delusive Adversaries with Adversarial Training (NeurIPS 2021 · similarity 0.86)
- Adversarial Robustness Against the Union of Multiple Perturbation Models (ICML 2020 · similarity 0.86)
- Seasoning Model Soups for Robustness to Adversarial and Natural Distribution Shifts (CVPR 2023 · similarity 0.86)
- Towards Deep Learning Models Resistant to Adversarial Attacks (ICLR 2018 · similarity 0.86)
- Adversarial Purification with the Manifold Hypothesis (AAAI 2024 · similarity 0.85)
- Certified Causal Defense with Generalizable Robustness (AAAI 2025 · similarity 0.85)
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.
