Research map / Adversarial robustness and security / Certified and empirical robustness
Robustness-Congruent Adversarial Training for Secure Machine Learning Model Updates
Daniele Angioni, Luca Demetrio, Maura Pintor, Luca Oneto, Davide Anguita, Battista Biggio, Fabio Roli
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 old model with a more robust new one (naive update) | old CIFAR-10 robust model · CIFAR-10 | RNF rate: roughly 2% to 6% | Hurt |
| fine-tune with PCT instead of naive replacement | naive update · CIFAR-10 (C3, C5) | RNFs / robust error: RNFs 6.30 -> 26.10; robust error 40.65 -> 66.10 | Hurt |
| fine-tune with PCT instead of naive replacement | naive update · ImageNet (I4, I5) | RNFs: 2.54 -> 37.22 | Hurt |
| add adversarial training to PCT (PCAT) | PCT · CIFAR-10 (C3, C5) | RNFs: 26.10 -> 10.95 | Improved |
| use RCAT instead of PCAT | PCAT · CIFAR-10 (C3, C5) | RNFs: 10.95 -> 6.00 | Improved |
| use RCAT instead of naive replacement | naive update · CIFAR-10 mean | delta RNFs: +0.18 | Improved |
| use RCAT instead of naive replacement | naive update · ImageNet mean | delta NFs: +0.55 | Improved |
| use RCAT instead of naive replacement | naive update · CIFAR-10 mean | delta robust error: -0.29 | No clear effect |
| use PCAT instead of naive replacement | naive update · ImageNet mean | delta NFs: -0.17 | No clear effect |
| RCAT versus PCAT in sequential ImageNet updates | PCAT · ImageNet sequential (I4, I5) | NFs: 3.32 -> 2.42 | Improved |
| RCAT versus PCT in sequential ImageNet updates | PCT · ImageNet sequential (I4, I5) | RNFs: 27.22 -> 2.02 | Improved |
| RCAT versus PCAT on update (C2, C3) | PCAT · CIFAR-10 (C2, C3) | NFs + RNFs: RNFs 3.15 -> 5.05 | Hurt |
Closest papers by meaning
- Robust Models are less Over-Confident (NeurIPS 2022 · similarity 0.91)
- On the Vulnerability of Adversarially Trained Models Against Two-faced Attacks (ICLR 2024 · similarity 0.91)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.89)
- Precise Accuracy / Robustness Tradeoffs in Regression: Case of General Norms (ICML 2024 · similarity 0.88)
- On the Alignment between Fairness and Accuracy: from the Perspective of Adversarial Robustness (ICML 2025 · similarity 0.88)
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.
