Research map / Adversarial robustness and security / Certified and empirical robustness
DISCO: Adversarial Defense with Local Implicit Functions
Chih-Hui Ho, Nuno Vasconcelos
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 |
|---|---|---|---|
| apply DISCO local implicit purification to Cifar10 classifier | WRN28-10 · Cifar10 AutoAttack Linf | robust accuracy: 66.58 -> 85.56 | Improved |
| add DISCO on top of robust ImageNet ResNet50 models | Hadi et al. · ImageNet AutoAttack | robust accuracy: 34.96 -> 50.6 | Improved |
| increase ImageNet training classes for DISCO from 100 to 1000 (ablation) | DISCO · ImageNet | robust accuracy: 59.84 -> 68.2 | Improved |
| kernel size 3 instead of 5 in local implicit module (ablation) | DISCO · ImageNet | robust accuracy: 60.1 -> 68.2 | Improved |
| kernel size 3 instead of 1 (ablation) | DISCO · ImageNet | robust accuracy: 69.52 -> 68.2 | No clear effect |
| train DISCO with BIM attack pairs instead of PGD (ablation) | DISCO · Cifar10 AutoAttack | robust accuracy: 85.56 -> 84.92 | No clear effect |
| train DISCO with FGSM attack pairs instead of PGD (ablation) | DISCO · Cifar10 AutoAttack | robust accuracy: 85.56 -> 77.13 | Hurt |
| cascade more DISCO stages (K up to 5) (ablation) | DISCO · Cifar10 BPDA adaptive attack | robust accuracy: saturating at 57.77 for K=5 | Improved |
| randomize defense stages differing from attacker K (ablation) | cascade DISCO · Cifar10 BPDA | robust accuracy: 52 -> 59.8 | Improved |
| transfer DISCO trained on ImageNet to Cifar10 testing (ablation) | DISCO · Cifar10 AutoAttack | robust accuracy: 85.56 -> 74.3 | Mixed |
Closest papers by meaning
- All You Need Is RAW: Defending against Adversarial Attacks with Camera Image Pipelines (ECCV 2022 · similarity 0.88)
- One Man's Trash Is Another Man's Treasure: Resisting Adversarial Examples by Adversarial Examples (CVPR 2020 · similarity 0.88)
- Defense Against Adversarial Images Using Web-Scale Nearest-Neighbor Search (CVPR 2019 · similarity 0.87)
- PAD: Patch-Agnostic Defense against Adversarial Patch Attacks (CVPR 2024 · similarity 0.85)
- PubDef: Defending Against Transfer Attacks From Public Models (ICLR 2024 · similarity 0.85)
- Enhancing the Adversarial Robustness via Manifold Projection (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.
