Research map / Architectures and efficiency / Pruning and efficient tuning
Learning Adversarially Robust Sparse Networks via Weight Reparameterization
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 |
|---|---|---|---|
| Use Twin-Rep weight reparameterization during adversarial training and pruning | VGG-16 and WideResNet-28-4 · CIFAR-10, 99% sparsity, AA∞ | robust accuracy: 37.4 -> 42.1 | Helped |
| Use Twin-Rep weight reparameterization during adversarial training and pruning | VGG-16 · SVHN, 99% sparsity, AA∞ | robust accuracy: 40.8 -> 44.2 | Helped |
| Use Twin-Rep weight reparameterization during adversarial training and pruning | WideResNet-28-4 · SVHN, 90% sparsity, AA∞ | robust accuracy: 47.0 -> 53.9 | Helped |
| Use Twin-Rep weight reparameterization during adversarial training and pruning | WideResNet-28-4 · SVHN, 99% sparsity, AA∞ | robust accuracy: 45.8 vs 45.8 | No clear effect |
| Use Twin-Rep instead of pruning original weights | ResNet-18 · CIFAR-10, unstructured pruning, 99% sparsity | robust accuracy: 40.5 -> 47.1 | Helped |
| Use Channel-Rep instead of pruning original weights | ResNet-18 · CIFAR-10, structured pruning, 70% sparsity | robust accuracy: 37.1 -> 47.7 | Helped |
| Use Twin-Rep instead of pruning original weights | ResNet-18 · CIFAR-100, unstructured pruning, 99% sparsity | robust accuracy: 20.6 -> 24.9 | Helped |
| Use Channel-Rep instead of pruning original weights | ResNet-18 · Tiny-ImageNet, structured pruning, 70% sparsity | robust accuracy: 12.0 -> 15.6 | Helped |
| Add Twin-Rep weight reparameterization (ablation) | ResNet-18 · CIFAR-100, unstructured pruning, 99% sparsity | robust accuracy: 20.6 -> 24.9 | Helped |
| Add Channel-Rep weight reparameterization (ablation) | ResNet-18 · CIFAR-10, structured pruning, 70% sparsity | robust accuracy: 37.1 -> 47.7 | Helped |
| Fuse Twin-Rep factors before adversarial training rather than maintain the Twin-Rep structure during training (ablation) | Twin-Rep · Weight-distribution and performance evaluation (Appendix C) | robust accuracy | Hurt |
Closest papers by meaning
- Sparsity Winning Twice: Better Robust Generalization from More Efficient Training (ICLR 2022 · similarity 0.90)
- Training Adversarially Robust Sparse Networks via Bayesian Connectivity Sampling (ICML 2021 · similarity 0.89)
- TWINS: A Fine-Tuning Framework for Improved Transferability of Adversarial Robustness and Generalization (CVPR 2023 · similarity 0.89)
- Powerpropagation: A sparsity inducing weight reparameterisation (NeurIPS 2021 · similarity 0.88)
- Adaptive Sharpness-Aware Pruning for Robust Sparse Networks (ICLR 2024 · 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.