Research map / Learning theory and algorithms / Kernels and regression theory
Training-Free Determination of Network Width via Neural Tangent Kernel
Tatsumi Sunada, Toshihiko Yamasaki, Atsuto Maki
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 smallest NTK eigenvalue saturation at initialization to pick width instead of training sweeps | DNN, CNN, ResNet width selection · Diabetes, California Housing, MNIST, CIFAR-10 (regression) | agreement of predicted width with test-loss saturation | Improved |
| vary optimizer, learning rate and initialization (ablation) | cardinal width on DNN · UCI Diabetes | cardinal width | No clear effect |
| compute mu_min on a subsample of the data (ablation) | cardinal width estimate · DNN on California Housing | relative saturation across widths | No clear effect |
| use the method outside lazy training (feature-learning regime) | cardinal width · experiments without lazy-training constraints | mu_min saturation vs test-loss saturation | No clear effect |
| compare test loss to NTK bound mu_min^-2 | two-layer ReLU on synthetic regression · make_regression, widths 16 to 1024 | test loss vs mu_min | Mixed |
Closest papers by meaning
- Fast Finite Width Neural Tangent Kernel (ICML 2022 · similarity 0.91)
- Finite Depth and Width Corrections to the Neural Tangent Kernel (ICLR 2020 · similarity 0.89)
- Finite-Width Neural Tangent Kernels from Feynman Diagrams (ICML 2026 · similarity 0.88)
- Tight Bounds on the Smallest Eigenvalue of the Neural Tangent Kernel for Deep ReLU Networks (ICML 2021 · similarity 0.88)
- The Surprising Effectiveness of Infinite-Width NTKs for Characterizing and Improving Model Training (AAAI 2025 · similarity 0.87)
- On the Impacts of the Random Initialization in the Neural Tangent Kernel Theory (NeurIPS 2024 · 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.
