Research map / Adversarial robustness and security / Certified and empirical robustness
Domain Knowledge Alleviates Adversarial Attacks in Multi-Label Classifiers
Stefano Melacci, Gabriele Ciravegna, Angelo Sotgiu, Ambra Demontis, Battista Biggio, Marco Gori, 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 |
|---|---|---|---|
| add domain-knowledge constraint loss (+C) during training (ablation) | FT · CIFAR-100 clean | F1 (%): 59.3 -> 64.0 | Improved |
| add domain-knowledge constraint loss (+C) (ablation) | TL · CIFAR-100 clean | F1 (%): 52.0 -> 55.1 | Improved |
| add domain-knowledge constraint loss (+C) (ablation) | TL · ANIMALS clean | F1 (%): 98.3 -> 98.6 | Improved |
| use strongly constrained training (+CC) instead of +C (ablation) | TL · PASCAL-Part clean | F1 (%): 70.0 -> 69.4 | Hurt |
| add knowledge constraints (+C) against white-box MKA (ablation) | FT · ANIMALS (eps=0.5) | quality under attack (%): 25.6 -> 31.7 | Improved |
| add knowledge constraints (+C) against black-box MKA (ablation) | FT · ANIMALS (eps=0.5) | quality under attack (%): 51.2 -> 76.7 | Improved |
| add knowledge-based rejection (Rej) (ablation) | FT · ANIMALS white-box MKA | quality under attack (%): 25.6 -> 39.3 | Improved |
| combine constraints and rejection (+C with Rej) (ablation) | FT · ANIMALS white-box MKA | quality under attack (%): 25.6 -> 60.7 | Improved |
| add knowledge-based rejection (Rej) (ablation) | FT · CIFAR-100 white-box MKA | quality under attack (%): 29.0 -> 31.1 | Improved |
| add knowledge-based rejection (Rej) (ablation) | FT · ANIMALS clean (eps=0) | quality (%): 98.6 -> 92.7 | Hurt |
| use knowledge-based rejection with constrained model instead of Neural Rejection (ablation) | FT · ANIMALS white-box MKA | quality under attack (%): 37.3 -> 60.7 | Improved |
| use knowledge-based rejection instead of Neural Rejection on unconstrained model (ablation) | TL · ANIMALS white-box MKA | quality under attack (%): 55.5 -> 49.8 | Hurt |
Closest papers by meaning
- Semantic-Aware Multi-Label Adversarial Attacks (CVPR 2024 · similarity 0.90)
- Characterizing the Evasion Attackability of Multi-label Classifiers (AAAI 2021 · similarity 0.88)
- MultiGuard: Provably Robust Multi-label Classification against Adversarial Examples (NeurIPS 2022 · similarity 0.88)
- Soften to Defend: Towards Adversarial Robustness via Self-Guided Label Refinement (CVPR 2024 · similarity 0.86)
- PatchDEMUX: A Certifiably Robust Framework for Multi-label Classifiers Against Adversarial Patches (CVPR 2025 · similarity 0.86)
- Adversarial Training for Defense Against Label Poisoning Attacks (ICLR 2025 · similarity 0.86)
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.
