Research map / Adversarial robustness and security / Certified and empirical robustness
Theoretically Principled Trade-off between Robustness and Accuracy
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 TRADES instead of Madry robust optimization | CIFAR10, WRN-34-10 · CIFAR10, FGSM^20 (PGD) white-box, ε=0.031 (ℓ∞) | robust accuracy: 47.04% -> 56.61% | Helped |
| Increase TRADES regularization from 1/λ=1 to 1/λ=6 (ablation) | CIFAR10, WRN-34-10 · CIFAR10, FGSM^20 (PGD) white-box | robust accuracy: 49.14% -> 56.61% | Helped |
| Increase TRADES regularization from 1/λ=1 to 1/λ=6 (ablation) | CIFAR10, WRN-34-10 · CIFAR10, FGSM^20 (PGD) white-box | natural accuracy: 88.64% -> 84.92% | Hurt |
| Increase regularization 1/λ (ablation) | TRADES on CIFAR10 · CIFAR10, FGSM^20 white-box, ε=0.031 (ℓ∞) | robust accuracy: 26.53 ± 1.1698 -> 50.64 ± 0.3336 | Helped |
| Increase regularization 1/λ (ablation) | TRADES on CIFAR10 · CIFAR10, FGSM^20 white-box, ε=0.031 (ℓ∞) | natural accuracy: 91.31 ± 0.0579 -> 81.72 ± 0.0286 | Hurt |
| Increase regularization 1/λ (ablation) | TRADES on MNIST · MNIST, FGSM^40 white-box, ε=0.1 | robust accuracy: 91.09 ± 0.0385 -> 95.65 ± 0.0340 | Helped |
| Increase regularization 1/λ (ablation) | TRADES on MNIST · MNIST, FGSM^40 white-box, ε=0.1 | natural accuracy: 99.41 ± 0.0235 -> 99.16 ± 0.0403 | No clear effect |
| Use TRADES rather than Madry | MNIST CNN · MNIST, FGSM^40 black-box (Natural source) | robust accuracy: 97.43% -> 97.63% | Helped |
| Use TRADES rather than Madry | CIFAR10 WRN-34-10 · CIFAR10, FGSM^20 black-box (Madry source) | robust accuracy: 66.00% -> 70.14% | Helped |
| Use TRADES rather than Madry robust optimization | MNIST CNN · MNIST, FGSM^40 (PGD) white-box, ε=0.3 (ℓ∞) | robust accuracy: 96.01% -> 96.07% | No clear effect |
Closest papers by meaning
- Certified Adversarial Robustness with Additive Noise (NeurIPS 2019 · similarity 0.92)
- Expressive Losses for Verified Robustness via Convex Combinations (ICLR 2024 · similarity 0.91)
- Fantastic Robustness Measures: The Secrets of Robust Generalization (NeurIPS 2023 · similarity 0.91)
- Precise Accuracy / Robustness Tradeoffs in Regression: Case of General Norms (ICML 2024 · similarity 0.91)
- On the Scalability of Certified Adversarial Robustness with Generated Data (NeurIPS 2024 · similarity 0.89)
- Provably Cost-Sensitive Adversarial Defense via Randomized Smoothing (ICML 2025 · similarity 0.89)
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.