Research map / Adversarial robustness and security / Certified and empirical robustness
Learning Extremely Lightweight and Robust Model with Differentiable Constraints on Sparsity and Condition Number
Xian Wei, Yangyu Xu, Yanhui Huang, Hairong Lv, Hai Lan, Mingsong Chen, Xuan Tang
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 condition number and sparsity constraints with tensor product (ARLST) vs ATMC | VGG-16 · CIFAR-10 PGD, 200x | robust accuracy: 36.1 -> 37.5 | Improved |
| ARLST instead of HYDRA at 200x compression | VGG-16 · CIFAR-10 PGD, 200x | robust accuracy: 35.5 -> 37.5 | Improved |
| ARLST instead of ADMM at 200x compression | VGG-16 · CIFAR-10 PGD, 200x | robust accuracy: 30.5 -> 37.5 | Improved |
| ARLST instead of ATMC at 100x compression | VGG-16 · CIFAR-10 PGD, 100x | robust accuracy: 41.1 -> 41.3 | Improved |
| ARLST instead of HYDRA at low compression (10x) | VGG-16 · CIFAR-10 PGD, 10x | robust accuracy: 46.2 -> 46.3 | No clear effect |
| ARLST instead of ATMC on ImageNet-1K 10x | VGG-16 · ImageNet-1K, 10x | robust accuracy: 27.8 -> 28.7 | Improved |
| ARLST instead of ATMC under PGD on CIFAR-100 (100x) | VGG-16 · CIFAR-100 PGD, 100x | robust accuracy: 21.3 -> 22.1 | Improved |
| ARLST instead of ATMC under AutoAttack on CIFAR-100 (100x) | VGG-16 · CIFAR-100 AutoAttack, 100x | robust accuracy: 19.9 -> 20.8 | Improved |
| ARLST instead of HYDRA on CCT at 10x compression | CCT · CIFAR-10 FGSM, 10x | robust accuracy: 36.45 -> 56.98 | Improved |
| ARLST instead of HYDRA on CCT at 5x compression | CCT · CIFAR-10 FGSM, 5x | robust accuracy: 55.54 -> 61.04 | Improved |
| add sparsity regularizer g (mu3 = 0.00001) to separable matrices (ablation) | ARLST VGG-16 · CIFAR-10 (last FC layer weights) | weight sparsity | Improved |
| set condition number weights mu1 = 0.0001 and mu2 = 1 (ablation) | ARLST · CIFAR-10 PGD | robust accuracy | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- HYDRA — HYDRA: Pruning Adversarially Robust Neural Networks (NeurIPS 2020) · better on 11 of 11 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Adversarial Robustness vs. Model Compression, or Both? (ICCV 2019 · similarity 0.90)
- Robust low-rank training via approximate orthonormal constraints (NeurIPS 2023 · similarity 0.89)
- Learning Adversarially Robust Sparse Networks via Weight Reparameterization (AAAI 2023 · similarity 0.88)
- Fast is better than free: Revisiting adversarial training (ICLR 2020 · similarity 0.87)
- Dynamical Low-Rank Compression of Neural Networks with Robustness under Adversarial Attacks (NeurIPS 2025 · similarity 0.87)
- Towards Certificated Model Robustness Against Weight Perturbations (AAAI 2020 · 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.
