Research map / Adversarial robustness and security / Certified and empirical robustness
DSRNA: Differentiable Search of Robust Neural Architectures
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 |
|---|---|---|---|
| Optimize architecture variables to maximize a certified robustness lower bound alongside validation accuracy | Differentiable NAS (DSRNA-CB) · CIFAR-10, PGD (100) | accuracy (%): 56.83 ± 0.21 -> 59.93 ± 0.24 | Helped |
| Optimize architecture variables to maximize a Jacobian norm robustness bound alongside validation accuracy | Differentiable NAS (DSRNA-Jacobian) · CIFAR-10, FGSM | accuracy (%): 65.29 ± 0.03 -> 68.92 ± 0.02 | Helped |
| Use certified-bound robustness optimization instead of adversarially trained architecture variables | DSRNA-CB · ImageNet, PGD (100) | accuracy (%): 46.59 ± 0.15 -> 45.39 ± 0.18 | Hurt |
| Use certified-bound robustness optimization instead of adversarially trained architecture variables | DSRNA-CB · ImageNet, C&W | accuracy (%): 36.69 ± 0.09 -> 43.64 ± 0.19 | Helped |
| Use Jacobian norm robustness optimization rather than adversarially trained architecture variables | DSRNA-Jacobian · MNIST, C&W | accuracy (%): 78.93 ± 0.03 -> 95.37 ± 0.02 | Helped |
| Use DSRNA-CB rather than DSRNA-Jacobian | DSRNA · CIFAR-10, PGD (100) | accuracy (%): 59.47 ± 0.14 -> 59.93 ± 0.24 | Helped |
| Use DSRNA-CB rather than DSRNA-Jacobian | DSRNA · CIFAR-10 architecture search | search time: 0.4 GPU days -> 4 GPU days | Hurt |
| Use DSRNA-CB rather than DSRNA-Jacobian | DSRNA · ImageNet, PGD (100) | accuracy (%): 43.79 ± 0.11 -> 45.39 ± 0.18 | Helped |
| Use DSRNA-Jacobian rather than DSRNA-CB | DSRNA · MNIST, C&W | accuracy (%): 94.02 ± 0.08 -> 95.37 ± 0.02 | Helped |
| Optimize architecture robustness directly rather than relying on adversarial training or random noise | DSRNA · CIFAR-10, C&W | accuracy (%): 42.58 ± 0.04 -> 63.01 ± 0.07 | Helped |
Closest papers by meaning
- Towards Accurate and Robust Architectures via Neural Architecture Search (CVPR 2024 · similarity 0.91)
- Robust NAS under adversarial training: benchmark, theory, and beyond (ICLR 2024 · similarity 0.91)
- Neural Architecture Search for Wide Spectrum Adversarial Robustness (AAAI 2023 · similarity 0.91)
- Neural Architecture Design and Robustness: A Dataset (ICLR 2023 · similarity 0.91)
- When NAS Meets Robustness: In Search of Robust Architectures Against Adversarial Attacks (CVPR 2020 · similarity 0.90)
- Accelerating Adversarially Robust Model Selection for Deep Neural Networks via Racing (AAAI 2024 · 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.