Research map / Adversarial robustness and security / Certified and empirical robustness
Exploring Memorization in Adversarial Training
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 temporal ensembling to PGD-AT | PGD-AT · CIFAR-10 | PGD-10 best accuracy: 52.64 -> 55.79 | Helped |
| Add temporal ensembling to PGD-AT | PGD-AT · CIFAR-100 | AutoAttack best accuracy: 24.72 -> 26.30 | Helped |
| Add temporal ensembling to PGD-AT | PGD-AT · SVHN | AutoAttack best accuracy: 46.61 -> 51.44 | Helped |
| Add temporal ensembling to TRADES | TRADES · CIFAR-10 | PGD-10 best accuracy: 53.32 -> 55.15 | Helped |
| Add temporal ensembling to TRADES | TRADES · SVHN | AutoAttack best accuracy: 40.36 -> 51.45 | Helped |
| Add temporal ensembling to PGD-AT | PGD-AT · CIFAR-10 | PGD-10 final accuracy: 44.92 -> 54.83 | Helped |
| Use a clean cross-entropy term with a gradual increase of the adversarial-loss weight (ablation) | PGD-AT on random-label CIFAR-10 · CIFAR-10 random labels | training convergence | Helped |
| Increase the perturbation budget from a small value to at least 2/255 (ablation) | PGD-AT on random-label CIFAR-10 · CIFAR-10 random labels | training convergence | Hurt |
| Add data augmentation, weight decay, or dropout (ablation) | TRADES on random-label CIFAR-10 · CIFAR-10 random labels | robust training accuracy | No clear effect |
| Increase label-noise rate (ablation) | Adversarial training · CIFAR-10 | generalization gap | Hurt |
| Use a larger perturbation budget (ablation) | PGD-AT with true labels · CIFAR-10 | robust training convergence | Hurt |
| Filter training examples using co-teaching (ablation) | PGD-AT · CIFAR-10 | AutoAttack best accuracy: 50.21 -> 49.60 | Hurt |
Closest papers by meaning
- On the Over-Memorization During Natural, Robust and Catastrophic Overfitting (ICLR 2024 · similarity 0.90)
- The Effectiveness of Random Forgetting for Robust Generalization (ICLR 2024 · similarity 0.89)
- Soften to Defend: Towards Adversarial Robustness via Self-Guided Label Refinement (CVPR 2024 · similarity 0.89)
- Rethinking LLM Memorization through the Lens of Adversarial Compression (NeurIPS 2024 · similarity 0.87)
- To Grok or not to Grok: Disentangling Generalization and Memorization on Corrupted Algorithmic Datasets (ICLR 2024 · similarity 0.87)
- Towards Memorization Estimation: Fast, Formal and Free (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.