Research map / Architectures and efficiency / CNN design
On Exact Computation with an Infinitely Wide Neural Net
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 |
|---|---|---|---|
| Add global average pooling to the vanilla convolutional architecture | CNN and CNTK · CIFAR-10 | classification accuracy: The paper reports an increase of 8% - 10% for both CNN and CNTK. | Helped |
| Increase depth | CNN and CNTK · CIFAR-10 | classification accuracy | Mixed |
| Use an exact CNTK rather than random-feature approximation | Vanilla CNTK · CIFAR-2 | classification accuracy: 87.25% -> 92.15%; 87.80% -> 93.05%; 85.35% -> 91.95% | Helped |
| Use an exact CNTK rather than random-feature approximation | CNTK-GAP · CIFAR-2 | classification accuracy: 51.10% -> 71.05%; 56.55% -> 95.40%; 90.65% -> 95.70% | Helped |
| Use a finite-width CNN instead of its infinite-width CNTK | CNN and corresponding CNTK · CIFAR-10 | classification accuracy: For depth 11 with GAP: 77.43% -> 82.92%; for depth 3 with GAP: 70.47% -> 63.81% | Mixed |
Closest papers by meaning
- Harnessing the Power of Infinitely Wide Deep Nets on Small-data Tasks (ICLR 2020 · similarity 0.91)
- The Surprising Effectiveness of Infinite-Width NTKs for Characterizing and Improving Model Training (AAAI 2025 · similarity 0.87)
- Just One Layer Norm Guarantees Stable Extrapolation (NeurIPS 2025 · similarity 0.86)
- An Infinite-Width Analysis on the Jacobian-Regularised Training of a Neural Network (ICML 2024 · similarity 0.86)
- On the Impacts of the Random Initialization in the Neural Tangent Kernel Theory (NeurIPS 2024 · similarity 0.85)
- Global Convergence and Rich Feature Learning in $L$-Layer Infinite-Width Neural Networks under $\mu$ Parametrization (ICML 2025 · similarity 0.85)
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.