Research map / Label-efficient and robust learning / Noisy labels
The Curious Case of Benign Memorization
Sotiris Anagnostidis, Gregor Bachmann, Lorenzo Noci, Thomas Hofmann
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 standard data augmentation during training on random labels | ResNet18 · CIFAR10 | K-NN probing accuracy (%): 12.4 -> 76.2 | Improved |
| Add standard data augmentation during training on random labels | VGG11 · CIFAR10 | K-NN probing accuracy (%): 15.0 -> 71.9 | Improved |
| Add standard data augmentation during training on random labels | ResNet18 · CIFAR100 | K-NN probing accuracy (%): 5.0 -> 42.7 | Improved |
| Add standard data augmentation during training on random labels | VGG11 · CIFAR100 | K-NN probing accuracy (%): 6.0 -> 46.4 | Improved |
| Add standard data augmentation during training on random labels | ResNet18 · TinyImageNet | K-NN probing accuracy (%): 2.4 -> 33.4 | Improved |
| Add standard data augmentation during training on random labels | VGG11 · TinyImageNet | K-NN probing accuracy (%): 1.2 -> 29.4 | Improved |
| Use independent same-class samples as augmentations instead of correlated standard augmentations (ablation) | ResNet18 trained on random labels · CIFAR10 | K-NN probing accuracy | Hurt |
| Increase the number of label-preserving augmentations (ablation) | ResNet18 trained on random labels · CIFAR10 | K-NN probing accuracy | Improved |
| Increase augmentation strength (ablation) | ResNet18 trained on random labels · Not specified | K-NN probing accuracy | Improved |
| Increase the number of random classes (ablation) | ResNet18 trained on random labels · Not specified | K-NN probing accuracy | Improved |
| Increase projector hidden-layer size (ablation) | ResNet18 trained on random labels · Not specified | K-NN probing accuracy | Improved |
Closest papers by meaning
- A Closer Look at Memorization in Deep Networks (ICML 2017 · similarity 0.91)
- How Spurious Features are Memorized: Precise Analysis for Random and NTK Features (ICML 2024 · similarity 0.90)
- The Pitfalls of Memorization: When Memorization Hurts Generalization (ICLR 2025 · similarity 0.90)
- Unveiling Memorization–Generalization Coexistence: A Case Study on Arithmetic Tasks with Label Noise (NeurIPS 2026 · similarity 0.89)
- On the Over-Memorization During Natural, Robust and Catastrophic Overfitting (ICLR 2024 · similarity 0.89)
- Towards Memorization Estimation: Fast, Formal and Free (ICML 2025 · similarity 0.88)
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.
