Research map / Adversarial robustness and security / Certified and empirical robustness
Towards Better Robust Generalization with Shift Consistency Regularization
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 |
|---|---|---|---|
| add shift consistency regularization (SCR) to FS adversarial training | FS on CIFAR-10 · CIFAR-10 PGD100 (eps=8) | robust accuracy: 70.30 -> 71.60 is a gain; paper notes FS-SCR slightly worse than FS on PGD100 per text | Hurt |
| add SCR to FS adversarial training | FS on CIFAR-10 · CIFAR-10 PGD20 (eps=8) | robust accuracy: 70.50 -> 76.45 | Helped |
| add SCR to FS adversarial training | FS on CIFAR-100 · CIFAR-100 PGD20 (eps=8) | robust accuracy: 47.20 -> 48.87 | Helped |
| add SCR to FS adversarial training | FS on SVHN · SVHN PGD20 (eps=8) | robust accuracy: 62.90 -> 70.24 | Helped |
| adding SCR to AT, TRADES, and FS | AT, TRADES, FS on CIFAR-10 · CIFAR-10 FGSM/PGD/CW/AA attacks | robust accuracy: consistent gains across baselines | Helped |
| use SIC loss only (drop CE term) under adaptive attack on FS+SCR (ablation) | FS+SCR on CIFAR-10 · CIFAR-10 adaptive PGD20 | robust accuracy: 80.69 (0*CE+1*SIC) vs 79.19 (1*CE+1*SIC) | Mixed |
| increase SIC weight relative to CE in FS+SCR (ablation) | FS+SCR on CIFAR-100 · CIFAR-100 adaptive PGD20 | robust accuracy: 48.96 at 1*CE+1*SIC vs 34.65 at 1*CE+0.1*SIC under CW20 | Mixed |
Closest papers by meaning
- Consistency Regularization for Adversarial Robustness (AAAI 2022 · similarity 0.91)
- Fantastic Robustness Measures: The Secrets of Robust Generalization (NeurIPS 2023 · similarity 0.90)
- Rethinking Invariance Regularization in Adversarial Training to Improve Robustness-Accuracy Trade-off (ICLR 2025 · similarity 0.90)
- Stability and Generalization of Adversarial Training for Shallow Neural Networks with Smooth Activation (NeurIPS 2024 · similarity 0.88)
- Kernel Learning with Adversarial Features: Numerical Efficiency and Adaptive Regularization (NeurIPS 2025 · similarity 0.88)
- A Closer Look at Curriculum Adversarial Training: From an Online Perspective (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.