Research map / Adversarial robustness and security / Certified and empirical robustness
Robustness and Accuracy Could Be Reconcilable by (Proper) Definition
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 |
|---|---|---|---|
| Replace KL divergence with squared error in PGD-AT | PGD-AT on CIFAR-10 · CIFAR-10, AutoAttack (ℓ∞, ε=8/255) | accuracy (%): 48.39 $\pm$ 0.14 -> 49.41 $\pm$ 0.27 | Helped |
| Add loss clipping to squared-error PGD-AT (ablation) | PGD-AT on CIFAR-10 · CIFAR-10, clean | accuracy (%): 82.13 $\pm$ 0.14 -> 82.80 $\pm$ 0.16 | Helped |
| Add loss clipping to squared-error PGD-AT (ablation) | PGD-AT on CIFAR-10 · CIFAR-10, AutoAttack (ℓ∞, ε=8/255) | accuracy (%): 49.41 $\pm$ 0.27 -> 49.63 $\pm$ 0.17 | Helped |
| Replace KL divergence with squared error in TRADES | TRADES on CIFAR-10 · CIFAR-10, clean | accuracy (%): 81.47 $\pm$ 0.12 -> 83.50 $\pm$ 0.05 | Helped |
| Replace KL divergence with squared error in TRADES | TRADES on CIFAR-10 · CIFAR-10, AutoAttack (ℓ∞, ε=8/255) | accuracy (%): 49.14 $\pm$ 0.16 -> 49.44 $\pm$ 0.35 | Helped |
| Add loss clipping to squared-error TRADES (ablation) | TRADES on CIFAR-10 · CIFAR-10, clean | accuracy (%): 83.50 $\pm$ 0.05 -> 83.75 $\pm$ 0.14 | Helped |
| Add loss clipping to squared-error TRADES (ablation) | TRADES on CIFAR-10 · CIFAR-10, AutoAttack (ℓ∞, ε=8/255) | accuracy (%): 49.44 $\pm$ 0.35 -> 49.57 $\pm$ 0.28 | Helped |
| Use squared error rather than KL and tune TRADES β to 3 | TRADES with 1M DDPM data · CIFAR-10, clean | accuracy (%): 85.97 -> 88.61 | Helped |
| Use squared error rather than KL and tune TRADES β to 3 | TRADES with 1M DDPM data · CIFAR-10, AutoAttack (ℓ∞, ε=8/255) | accuracy (%): 60.73 -> 61.04 | No clear effect |
| Use squared error rather than KL and tune TRADES β to 4 | TRADES with 1M DDPM data · CIFAR-100, clean | accuracy (%): 59.18 -> 62.08 | Helped |
| Use squared error rather than KL and tune TRADES β to 4 | TRADES with 1M DDPM data · CIFAR-100, AutoAttack (ℓ∞, ε=8/255) | accuracy (%): 30.81 -> 31.40 | Helped |
| Reduce TRADES β from 6 to 2 (ablation) | Squared-error TRADES with DDPM data · CIFAR-10, AutoAttack (ℓ∞, ε=8/255) | accuracy (%): 60.78 $\pm$ 0.16 -> 60.13 $\pm$ 0.21 | Hurt |
Closest papers by meaning
- Theoretically Principled Trade-off between Robustness and Accuracy (ICML 2019 · similarity 0.91)
- GREAT Score: Global Robustness Evaluation of Adversarial Perturbation using Generative Models (NeurIPS 2024 · similarity 0.89)
- Precise Accuracy / Robustness Tradeoffs in Regression: Case of General Norms (ICML 2024 · similarity 0.89)
- OODRobustBench: a Benchmark and Large-Scale Analysis of Adversarial Robustness under Distribution Shift (ICML 2024 · similarity 0.88)
- Adversarial Robustness Limits via Scaling-Law and Human-Alignment Studies (ICML 2024 · similarity 0.88)
- On the Alignment between Fairness and Accuracy: from the Perspective of Adversarial Robustness (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.