Research map / Adversarial robustness and security / Black-box and transfer attacks
When NAS Meets Robustness: In Search of Robust Architectures Against Adversarial Attacks
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 a densely connected architecture | Adversarially trained cell-based networks · CIFAR-10 | PGD adversarial accuracy | Helped |
| Add convolution operations to direct edges rather than allocating them to skip edges | Cell-based networks under small and medium computational budgets · CIFAR-10 | adversarial accuracy | Helped |
| Increase the number of convolution operations | Cell-based networks · CIFAR-10 | adversarial accuracy | Helped |
| Increase model size while keeping the RobNet-large cell architecture | RobNet-large · CIFAR-10 | PGD^100 accuracy: 49.24% -> 50.26% | Helped |
| Use cell-free search guided by FSP matrix loss instead of cell-based search | RobNet · CIFAR-10 white-box | PGD^100 accuracy: 49.24% -> 52.57% | Helped |
| Filter candidate networks using lower FSP matrix loss in deeper cells | Cell-free architecture search · Cell-free search space | gap between clean and adversarial accuracy | Helped |
| Add feature denoising blocks | RobNet-large · CIFAR-10 | PGD^100 accuracy: 49.24% -> 49.97% | Helped |
| Add feature denoising blocks | ResNet-18 · CIFAR-10 | PGD^100 accuracy: 45.10% -> 45.82% | Helped |
| Transfer the architecture searched on CIFAR-10 to another dataset | RobNet-large and RobNet-free · SVHN, CIFAR-100, and Tiny-ImageNet | PGD^100 accuracy | Helped |
| Increase the direct-convolution proportion | Cell-based networks with a large computational budget · CIFAR-10 | adversarial accuracy | No clear effect |
| Use more convolution operations on direct edges within small or medium budgets | Cell-based networks · Large computational budget | adversarial accuracy | No clear effect |
Closest papers by meaning
- Robust NAS under adversarial training: benchmark, theory, and beyond (ICLR 2024 · similarity 0.93)
- Exploring Architectural Ingredients of Adversarially Robust Deep Neural Networks (NeurIPS 2021 · similarity 0.92)
- Revisiting Residual Networks for Adversarial Robustness (CVPR 2023 · similarity 0.92)
- Neural Architecture Design and Robustness: A Dataset (ICLR 2023 · similarity 0.92)
- Towards Accurate and Robust Architectures via Neural Architecture Search (CVPR 2024 · similarity 0.91)
- ROME is Forged in Adversity: Robust Distilled Datasets via Information Bottleneck (ICML 2025 · 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.