Research map / Label-efficient and robust learning / Recognition and long-tail
Spectral Bias in Practice: The Role of Function Frequency in Generalization
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 |
|---|---|---|---|
| Train longer (ablation) | CIFAR-10 image classifiers · CIFAR-10 | within-class versus between-class frequency separation | Helped |
| Increase model width (ablation) | Wide-ResNet and Shake-Shake models · CIFAR-10 | learnability of high-frequency label noise | Helped |
| Increase the number of training examples (ablation) | CIFAR-10 classifiers · CIFAR-10 | within-class versus between-class frequency separation | Helped |
| Apply more effective data augmentation (ablation) | CIFAR-10 classifiers · CIFAR-10 | within-class versus between-class frequency separation | Helped |
| Use modest Mixup augmentation (ablation) | CIFAR-10 classifiers · CIFAR-10 | within-class frequency | Helped |
| Use excessively strong Mixup augmentation (ablation) | CIFAR-10 classifiers · CIFAR-10 | within-class frequency | Hurt |
| Add self-distillation (ablation) | CIFAR-10 student classifiers · CIFAR-10 | within-class frequency | Helped |
| Increase weight decay (ablation) | CIFAR-10 classifiers · CIFAR-10 | speed of learning high-frequency target functions | Hurt |
| Increase radial-wave label-noise frequency (ablation) | CIFAR-10 classifiers · CIFAR-10 | speed of fitting label noise | Hurt |
| Train longer (ablation) | ImageNet classifiers · ImageNet | within-class versus between-class frequency separation | Helped |
| Evaluate within-class paths in more internally diverse classes | Most ImageNet models · ImageNet | within-class function frequency | Helped |
| Relate class diversity to function frequency | CoAtNet models · ImageNet | correlation between class diversity and within-class frequency | No clear effect |
Closest papers by meaning
- On the Spectral Bias of Neural Networks (ICML 2019 · similarity 0.92)
- Addressing Spectral Bias of Deep Neural Networks by Multi-Grade Deep Learning (NeurIPS 2024 · similarity 0.91)
- An Analytical Theory of Spectral Bias in the Learning Dynamics of Diffusion Models (NeurIPS 2025 · similarity 0.87)
- Do We Always Need the Simplicity Bias? Looking for Optimal Inductive Biases in the Wild (CVPR 2025 · similarity 0.86)
- Tuning Frequency Bias of State Space Models (ICLR 2025 · similarity 0.86)
- Inductive Gradient Adjustment for Spectral Bias in Implicit Neural Representations (ICML 2025 · similarity 0.86)
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.