Research map / Adversarial robustness and security / Certified and empirical robustness
HYDRA: Pruning Adversarially Robust Neural Networks
Read the paperCheck 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 |
|---|---|---|---|
| make pruning aware of robust objective (HYDRA) instead of least-weight-magnitude | VGG-16 adversarial training · CIFAR-10, 99% pruning | robust accuracy (era): 34.1 -> 41.7 | Helped |
| HYDRA instead of Adv-LWM | CNN-small, CROWN-IBP · CIFAR-10, 95% pruning | vra-t: 40.3 -> 40.0 | No clear effect |
| HYDRA instead of Adv-LWM | CNN-small MixTrain · SVHN, 99% pruning | vra-m: 19.6 -> 19.6 | No clear effect |
| HYDRA instead of Adv-ADMM | VGG16 adversarial training · CIFAR-10, 99% pruning | robust accuracy (era): 34.1 -> 37.9 | Helped |
| HYDRA instead of Adv-ADMM | MobileNet-v2 · CIFAR-10, 99% pruning | robust accuracy: 10.0 -> 26.4 | Helped |
| use scaled initialization of importance scores instead of Kaiming-normal (ablation) | HYDRA pruning step · CIFAR-10 VGG16, 99% pruning | robust accuracy (era): 36.5 -> 41.7 | Helped |
| prune pre-trained net instead of training compact net from scratch (ablation) | VGG16 adversarial training · CIFAR-10, 99% pruning | robust accuracy (era): 24.6 -> 41.7 | Helped |
| add scaling factor (k=6) to score initialization (ablation) | HYDRA pruning step · CIFAR-10 VGG-16, 99% | benign accuracy: 57.9 -> 66.8 | Helped |
| add scaling factor (k=6) to score initialization (ablation) | HYDRA pruning step · CIFAR-10 VGG-16, 99% | robust accuracy (era): 31.6 -> 35.8 | Helped |
| use more training data in the pruning step (ablation) | HYDRA · CIFAR-10, 99% pruning | era | Helped |
| more pruning epochs (ablation) | HYDRA · CIFAR-10, 99% pruning | era | Helped |
Closest papers by meaning
- A Combination of Noise and Bilateral Filters Achieve Supralinear and Scalable Adversarial Robustness in CNNs (CVPR 2026 · similarity 0.89)
- Robust Binary Models by Pruning Randomly-initialized Networks (NeurIPS 2022 · similarity 0.88)
- Masking Adversarial Damage: Finding Adversarial Saliency for Robust and Sparse Network (CVPR 2022 · similarity 0.88)
- Robust NAS under adversarial training: benchmark, theory, and beyond (ICLR 2024 · similarity 0.88)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.87)
- SAFE: Finding Sparse and Flat Minima to Improve Pruning (ICML 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.