Research map / Optimization / Implicit bias and generalization
Local Signal Adaptivity: Provable Feature Learning in Neural Networks Beyond Kernels
Stefani Karp, Ezra Winston, Yuanzhi Li, Aarti Singh
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 |
|---|---|---|---|
| increase background noise intensity from 0 to max | NTK · CIFAR-10 with ImageNet noise | accuracy retained (%): 69.75 | Hurt |
| increase background noise intensity from 0 to max | WRN · CIFAR-10 with ImageNet noise | accuracy retained (%): 98.97 | No clear effect |
| increase noise on CIFAR-Vehicles distractors | NTK · CIFAR-Vehicles | accuracy retained (%): 76.37 | Hurt |
| increase ImageNet noise intensity | NTK · CIFAR-2 | accuracy retained (%): 92.81 | Mixed |
| increase Gaussian noise stdev from 0 to 2 | NTK · CIFAR-2 Gaussian | accuracy retained (%): 85.54 | Hurt |
| increase Gaussian noise stdev from 0 to 2 | WRN · CIFAR-2 Gaussian | accuracy retained (%): 96.83 | No clear effect |
| increase NTK width factor (50, 100) on MNIST with Gaussian noise | finite-width NTK · MNIST with noise | test accuracy | No clear effect |
| use infinite-width NTK instead of finite-width | NTK · MNIST-20k with Gaussian noise | test accuracy | No clear effect |
| zero-noise comparison of CNN and NTK | NTK · MNIST zero noise | test accuracy | No clear effect |
Closest papers by meaning
- From Kernels to Features: A Multi-Scale Adaptive Theory of Feature Learning (ICML 2025 · similarity 0.84)
- Dynamic Sparse Training versus Dense Training: The Unexpected Winner in Image Corruption Robustness (ICLR 2025 · similarity 0.84)
- How does Gradient Descent Learn Features --- A Local Analysis for Regularized Two-Layer Neural Networks (NeurIPS 2024 · similarity 0.84)
- Critical feature learning in deep neural networks (ICML 2024 · similarity 0.84)
- Implicit Compressibility of Overparametrized Neural Networks Trained with Heavy-Tailed SGD (ICML 2024 · similarity 0.84)
- LAN: Learning to Adapt Noise for Image Denoising (CVPR 2024 · similarity 0.83)
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.
