Research map / Adversarial robustness and security / Certified and empirical robustness
Improving Adversarial Robustness of Deep Equilibrium Models with Explicit Regulations Along the Neural Dynamics
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 test-time input entropy reduction along neural dynamics (Sec. 3.2) (ablation) | DEQ-Large trained with TRADES · CIFAR-10 ALL robustness | ALL robustness (%): 50.33 -> 50.52 | Helped |
| Add random intermediate-state loss during AT (Sec. 3.3) (ablation) | DEQ-Large trained with TRADES · CIFAR-10 ALL robustness | ALL robustness (%): 50.33 -> 49.64 (decrease under ALL; PGD-10 rises 50.46 -> 51.10) | Helped |
| Add both Sec. 3.2 and Sec. 3.3 together (ablation) | DEQ-Large trained with TRADES · CIFAR-10 ALL robustness | ALL robustness (%): 50.33 -> 51.50 | Helped |
| Add Sec. 3.2 entropy reduction to PGD-AT DEQ (ablation) | DEQ-Large trained with PGD-AT · CIFAR-10 ALL robustness | ALL robustness (%): 47.58 -> 49.31 | Helped |
| Add Sec. 3.3 random intermediate loss to PGD-AT DEQ (ablation) | DEQ-Large trained with PGD-AT · CIFAR-10 ALL robustness | ALL robustness (%): 47.58 -> 48.09 | Helped |
| Add both Sec. 3.2 and Sec. 3.3 to PGD-AT DEQ (ablation) | DEQ-Large trained with PGD-AT · CIFAR-10 ALL robustness | ALL robustness (%): 47.58 -> 43.79 (AA drops to 43.79 while PGD rises to 49.22) | Mixed |
| Test-time entropy reduction (Sec. 3.2) with R=20, Tf=2 vs varied settings (ablation) | TRADES + Sec. 3.3 DEQ · CIFAR-10 min PGD robustness over intermediate attacks | robust accuracy (%): Tf=1 to 8 lowers accuracy (e.g. R=10: 55.07 at Tf=1, 52.83 at Tf=8); larger R helps at Tf=2 | Mixed |
Closest papers by meaning
- Lyapunov-Stable Deep Equilibrium Models (AAAI 2024 · similarity 0.92)
- Separation and Bias of Deep Equilibrium Models on Expressivity and Learning Dynamics (NeurIPS 2024 · similarity 0.91)
- Positive Concave Deep Equilibrium Models (ICML 2024 · similarity 0.88)
- Understanding Representation of Deep Equilibrium Models from Neural Collapse Perspective (NeurIPS 2024 · similarity 0.88)
- Stability and Generalization of Adversarial Training for Shallow Neural Networks with Smooth Activation (NeurIPS 2024 · similarity 0.87)
- Robust Distillation via Untargeted and Targeted Intermediate Adversarial Samples (CVPR 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.