Research map / Adversarial robustness and security / Certified and empirical robustness
Adversarially Robust Neural Architectures
Minjing Dong, Yanxi Li, Yunhe Wang, Chang Xu
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 confidence learning of architecture parameters (ablation) | RACL · CIFAR-10 PGD-20 | adversarial accuracy: 54.58 -> 55.68 | Improved |
| add confidence learning of architecture parameters (ablation) | RACL · CIFAR-10 AutoAttack | adversarial accuracy: 48.89 -> 50.07 | Improved |
| add confidence learning of architecture parameters (ablation) | RACL · CIFAR-10 clean | natural accuracy: 84.35 -> 84.04 | Hurt |
| add gradient norm lower-bound constraint (ablation) | RACL · CIFAR-10 PGD-20 | adversarial accuracy: 53.87 -> 55.68 | Improved |
| search with RACL instead of random search (ablation) | RACL · CIFAR-10 PGD-20 | adversarial accuracy: 51.36 -> 55.68 | Improved |
| use rho=0.001 instead of rho=0.01 (eta=0.9) (ablation) | RACL · CIFAR-10 PGD-20 | adversarial accuracy: 55.66 -> 55.68 | Improved |
| use eta=0.9 instead of eta=0.7 (rho=0.001) (ablation) | RACL · CIFAR-10 PGD-20 | adversarial accuracy: 55.00 -> 55.68 | Improved |
| replace PC-DARTS cells with RACL cells | PC-DARTS · CIFAR-10 PGD-100 | adversarial accuracy: 52.36 -> 55.32 | Improved |
| replace RobNet-free with RACL | RobNet-free · CIFAR-10 PGD-100 | adversarial accuracy: 52.57 -> 55.32 | Improved |
| replace DARTS with RACL on CIFAR-100 | DARTS · CIFAR-100 AutoAttack | adversarial accuracy: 24.30 -> 25.55 | Improved |
| replace NASNet with RACL on Tiny-ImageNet | NASNet · Tiny-ImageNet CW | adversarial accuracy: 41.93 -> 42.99 | Improved |
| replace NASNet with RACL under transfer black-box attack | NASNet · CIFAR-10 transfer black-box PGD-20 | adversarial accuracy: 83.21 -> 83.45 | Improved |
Closest papers by meaning
- Exploring Architectural Ingredients of Adversarially Robust Deep Neural Networks (NeurIPS 2021 · similarity 0.93)
- When NAS Meets Robustness: In Search of Robust Architectures Against Adversarial Attacks (CVPR 2020 · similarity 0.92)
- Evaluating the Robustness of Neural Networks: An Extreme Value Theory Approach (ICLR 2018 · similarity 0.92)
- Robust NAS under adversarial training: benchmark, theory, and beyond (ICLR 2024 · similarity 0.90)
- Towards Accurate and Robust Architectures via Neural Architecture Search (CVPR 2024 · similarity 0.90)
- A Recipe for Improved Certifiable Robustness (ICLR 2024 · similarity 0.89)
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.
