Research map / Learning theory and algorithms / Kernels and regression theory
Towards a General Theory of Infinite-Width Limits of Neural Classifiers
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 |
|---|---|---|---|
| Use discrete-time MF scaling instead of NTK scaling | Single-hidden-layer network · CIFAR2 | finite-width training-dynamics approximation | Helped |
| Use NTK scaling instead of discrete-time MF scaling | Single-hidden-layer network · CIFAR2 | finite-width training-dynamics approximation | Hurt |
| Use default scaling with unscaled learning rates | Single-hidden-layer network · CIFAR2 | final test cross-entropy loss | Hurt |
| Use intermediate scaling with q_sigma=-3/4 instead of NTK scaling | Single-hidden-layer network · CIFAR2 | finite-width training-dynamics approximation | Hurt |
| Use MF scaling with GD for a network with more than two hidden layers | Multi-layer network · CIFAR2 | non-trivial infinite-width limit | Hurt |
| Train with RMSProp instead of GD under MF scaling | Multi-layer network · CIFAR2 | non-trivial infinite-width limit | Helped |
| Use NTK scaling instead of MF scaling for a multi-layer network | Multi-layer network · CIFAR2 | preservation of model-decomposition terms | Mixed |
Closest papers by meaning
- On the Surprising Effectiveness of Large Learning Rates under Standard Width Scaling (NeurIPS 2025 · similarity 0.90)
- An Infinite-Width Analysis on the Jacobian-Regularised Training of a Neural Network (ICML 2024 · similarity 0.89)
- Infinite-Width Limit of a Single Attention Layer: Analysis via Tensor Programs (NeurIPS 2025 · similarity 0.89)
- Sharper Guarantees for Learning Neural Network Classifiers with Gradient Methods (ICLR 2025 · similarity 0.89)
- Adaptive kernel predictors from feature-learning infinite limits of neural networks (ICML 2025 · similarity 0.88)
- Universal Consistency of Wide and Deep ReLU Neural Networks and Minimax Optimal Convergence Rates for Kolmogorov-Donoho Optimal Function Classes (ICML 2024 · 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.