Research map / Adversarial robustness and security / Certified and empirical robustness
Confound from All Sides, Distill with Resilience: Multi-Objective Adversarial Paths to Zero-Shot Robustness
Junhao Dong, Jiao Liu, Xinghua Qu, Yew-Soon Ong
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 |
|---|---|---|---|
| use MOO-adversaries instead of untargeted adversaries (ablation) | MOO-AD · ImageNet PGD | robust accuracy: 29.13 -> 36.58 | Improved |
| use MOO-adversaries instead of targeted adversaries (ablation) | MOO-AD · ImageNet PGD | robust accuracy: 25.88 -> 36.58 | Improved |
| use MOO-adversaries instead of untargeted adversaries (ablation) | MOO-AD · OOD datasets (SUN397, Flower102, CIFAR-100) PGD | robust accuracy: 16.87 -> 24.74 | Improved |
| use targeted instead of untargeted adversaries (ablation) | MOO-AD · OOD datasets PGD | robust accuracy: 16.87 -> 20.59 | Improved |
| use targeted instead of untargeted adversaries (ablation) | MOO-AD · ImageNet clean | accuracy: 53.42 -> 50.94 | Hurt |
| increase number of MOO-adversaries from 1 to 10 (ablation) | MOO-AD · 15-dataset average PGD | robust accuracy: 32.17 -> 35.70 | Improved |
| increase number of MOO-adversaries from 10 to 30 (ablation) | MOO-AD · 15-dataset average PGD | robust accuracy: 35.70 -> 36.11 | No clear effect |
| increase trade-off factor beta (ablation) | MOO-AD · 15-dataset average | clean vs robust accuracy | Mixed |
| add both MOPA and GDM components with MOO-adversaries (ablation) | MOO-AD · 15-dataset average | PGD robust accuracy: 32.37 -> 35.70 | Improved |
| distill with MOO-AD instead of FARE | ViT-B student · 15-dataset average PGD | robust accuracy: 30.94 -> 35.70 | Improved |
| distill with MOO-AD instead of FARE | ViT-B student · 15-dataset average clean | accuracy: 56.68 -> 58.96 | Improved |
| distill with MOO-AD instead of FARE at larger radius 4/255 | ViT-B student · 15-dataset average PGD eps 4/255 | robust accuracy: 12.42 -> 18.47 | Improved |
Closest papers by meaning
- Allies Teach Better Than Enemies: Inverse Adversaries for Robust Knowledge Distillation (TPAMI 2026 · similarity 0.92)
- Robust Distillation via Untargeted and Targeted Intermediate Adversarial Samples (CVPR 2024 · similarity 0.91)
- Improving Zero-Shot Adversarial Robustness in Vision-Language Models by Closed-form Alignment of Adversarial Path Simplices (ICML 2025 · similarity 0.90)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.89)
- Out of Thin Air: Exploring Data-Free Adversarial Robustness Distillation (AAAI 2024 · similarity 0.88)
- Impartial Adversarial Distillation: Addressing Biased Data-Free Knowledge Distillation via Adaptive Constrained Optimization (AAAI 2024 · 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.
