Research map / Learning theory and algorithms / Kernels and regression theory
Over-Alignment vs Over-Fitting: The Role of Feature Learning Strength in Generalization
Taesun Yeom, Taehyeok Ha, Jaeho Lee
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 |
|---|---|---|---|
| use an intermediate output multiplier (optimal FLS) instead of default scale (ablation) | ResNet50 · CIFAR-100 | peak test accuracy (%): 53.57 -> 59.76 | Improved |
| increase FLS beyond optimum (smaller output multiplier below optimum) (ablation) | VGG19, ResNet18/34/50 · CIFAR-100 | peak test accuracy | Hurt |
| decrease FLS (larger output multiplier) (ablation) | VGG19, ResNet18/34/50 · CIFAR-100 | peak test accuracy | Hurt |
| harder dataset (higher effective dimensionality 32 -> 128) raises the benefit of tuning FLS (ablation) | ResNet18 · BigGAN dog dataset | test accuracy gap to optimal FLS | Improved |
| stop training by minimum validation loss instead of peak test accuracy (ablation) | ResNet-18 · BigGAN dog dataset (edim=128) | optimal output multiplier | No clear effect |
| scale output multiplier with width and dataset size per theory (c* ~ n^-2 h^-1) | 5-layer vanilla CNN · BigGAN dataset (edim=128) | optimal output multiplier | Improved |
Closest papers by meaning
- Weak-to-Strong Generalization Even in Random Feature Networks, Provably (ICML 2025 · similarity 0.88)
- From Linear to Nonlinear: Provable Weak-to-Strong Generalization through Feature Learning (NeurIPS 2025 · similarity 0.88)
- Feature Contamination: Neural Networks Learn Uncorrelated Features and Fail to Generalize (ICML 2024 · similarity 0.87)
- From Kernels to Features: A Multi-Scale Adaptive Theory of Feature Learning (ICML 2025 · similarity 0.87)
- Gradient-Weight Alignment as a Train-Time Proxy for Generalization in Classification Tasks (NeurIPS 2025 · similarity 0.87)
- On the Mechanisms of Weak-to-Strong Generalization: A Theoretical Perspective (NeurIPS 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.
