Research map / Architectures and efficiency / CNN design
Neural Kernels Without Tangents
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 the Myrtle10 Gaussian compositional kernel instead of the NTK | kernel classifier · CIFAR-10 | accuracy: 81.4 -> 88.2 | Helped |
| Use the Myrtle10 CNN instead of its corresponding compositional kernel | Myrtle10 architecture · CIFAR-10 | accuracy: 88.2 -> 91.2 | Helped |
| Add flip augmentation (ablation) | Myrtle10 Gaussian kernel · CIFAR-10 | accuracy: 88.2 -> 89.8 | Helped |
| Add flip augmentation (ablation) | Myrtle10 CNN · CIFAR-10 | accuracy: 91.2 -> 93.4 | Helped |
| Add cutout and random crops to flip augmentation (ablation) | Myrtle10 CNN · CIFAR-10 | accuracy: 93.4 -> 96.0 | Helped |
| Use ZCA preprocessing instead of mean subtraction and standard deviation normalization (ablation) | Myrtle5 compositional kernel · CIFAR-10 | accuracy: 77.7 -> 85.8 | Helped |
| Use ZCA preprocessing instead of mean subtraction and standard deviation normalization (ablation) | Myrtle5 CNN · CIFAR-10 | accuracy: 87.8 -> 89.8 | Helped |
| Use a Gaussian embedding rather than the ReLU kernel (ablation) | Myrtle10 compositional kernel · CIFAR-10 | accuracy: 87.5 -> 88.2 | Helped |
| Add batch normalization (ablation) | Myrtle10 CNN · CIFAR-10.0 | accuracy: 64.7 -> 70.3 | Helped |
| Add flip augmentation (ablation) | Myrtle10 CNN · CIFAR-10.0 | accuracy: 64.7 -> 71.4 | Helped |
| Add batch normalization alongside flip augmentation (ablation) | Myrtle10 CNN · CIFAR-10.0 | accuracy: 71.4 -> 74.7 | Helped |
| Use the compositional kernel instead of the same-architecture CNN | Myrtle5 models · CIFAR-10 subsamples | accuracy: 17.22 vs 11.29 (training size 10); 60.46 vs 53.27 (training size 1280) | Mixed |
Closest papers by meaning
- Optimized Random Features for the Neural Tangent Kernel (Student Abstract) (AAAI 2025 · similarity 0.87)
- Neural Tangent Kernels Motivate Cross-Covariance Graphs in Neural Networks (ICML 2024 · similarity 0.86)
- The Surprising Effectiveness of Infinite-Width NTKs for Characterizing and Improving Model Training (AAAI 2025 · similarity 0.86)
- Learning Input Encodings for Kernel-Optimal Implicit Neural Representations (ICML 2025 · similarity 0.86)
- Scalable Neural Network Kernels (ICLR 2024 · similarity 0.86)
- Equivariant Neural Tangent Kernels (ICML 2025 · similarity 0.86)
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.