Research map / Adversarial robustness and security / Certified and empirical robustness
A3: A Platform of Searching for Robust Neural Architectures and Efficient Adversarial Attacks
Jialiang Sun, Wen Yao, Tingsong Jiang, Chao Li, Xiaoqian Chen
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 DE instead of random search to search the attack (ablation) | attack search on Regularization ResNet-18 · CIFAR10 l_inf | robust accuracy (%): 53.6 -> 5.00 | Improved |
| search attack with DE instead of using manual PGD-7 | PGDAT ResNet-50 · CIFAR10 l_inf | robust accuracy (%): 54.6 -> 50.8 | Improved |
| use multi-objective NSGA-II instead of DE for attack search (ablation) | PGDAT ResNet-50 · CIFAR10 l_inf | time cost (s): 458.9 -> 145.88 time, 50.8 -> 52.0 accuracy | Mixed |
| use NSGA-II instead of random search for attack search (ablation) | attack search · CIFAR10 l_inf and l_2 | robust accuracy (%) | Improved |
| use DE instead of random search for NAS (FGSM evaluation) (ablation) | NAS for robust architectures · CIFAR10 PGD-LinfAttack | robust accuracy (%): 43.6 -> 45.2 | Improved |
| use DE instead of random search for NAS (ablation) | NAS search cost · FGSM evaluation | search time (GPU days): 1.15 -> 12 | Hurt |
| use Jacobian instead of FGSM as NAS robustness metric (ablation) | DARTS · CIFAR10 PGD-LinfAttack | robust accuracy (%): 38.8 -> 51.2 | Improved |
| use Jacobian instead of FGSM as NAS robustness metric (ablation) | SmoothDARTS · CIFAR10 PGD-LinfAttack | robust accuracy (%): 53.0 -> 42.2 | Hurt |
| use searched attack AAA instead of FGSM as NAS robustness evaluation (ablation) | Random search NAS · CIFAR10 AAA | robust accuracy (%): 41.6 -> 49.4 | Improved |
| use more robust source model (FairDARTS_Jacobian) for attack search (ablation) | random search attack on SmoothDARTS_FGSM · CIFAR10 | robust accuracy (%): 54.0 -> 52.8 | Improved |
| use natural-noise evaluation in NAS (DARTS_Natural) vs manual ResNet-18 (ablation) | DARTS · CIFAR10-C snow | accuracy (%): 76.56 -> 79.17 | Improved |
| use SmoothDARTS with Natural evaluation instead of DE (ablation) | NAS · CIFAR10 clean | accuracy (%): 92.61 -> 89.42 | Hurt |
Closest papers by meaning
- Efficient Robustness Evaluation via Constraint Relaxation (AAAI 2025 · similarity 0.89)
- Towards Accurate and Robust Architectures via Neural Architecture Search (CVPR 2024 · similarity 0.89)
- Robust NAS under adversarial training: benchmark, theory, and beyond (ICLR 2024 · similarity 0.88)
- SuperDeepFool: a new fast and accurate minimal adversarial attack (NeurIPS 2024 · similarity 0.88)
- Rethinking the Adversarial Robustness of Multi-Exit Neural Networks in an Attack-Defense Game (CVPR 2025 · similarity 0.88)
- First Line of Defense: A Robust First Layer Mitigates Adversarial Attacks (AAAI 2025 · similarity 0.87)
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.
