Research map / Learning theory and algorithms / Kernels and regression theory
Scaling Neural Tangent Kernels via Sketching and Random Features
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 CNTKSketch with 16,384 features instead of exact CNTK | CNTK regression · CIFAR-10 | test accuracy (%): 70.47$^*$ -> {72.06} | Mixed |
| Use CNTKSketch with 16,384 features instead of GradRF with 42,816 features | CNTK regression · CIFAR-10 | test accuracy (%): 65.21 -> {72.06} | Helped |
| Replace exact CNTK computation with CNTKSketch | CNTK regression · CIFAR-10 | time (s): > 1,000,000 -> 5{,}160 | Helped |
| Use NTKSketch instead of RFF | kernel ridge regression · MillionSongs | MSE: 109.50 -> {92.83} | Helped |
| Use NTKSketch instead of NTKRF | kernel ridge regression · MillionSongs | MSE: 94.27 -> {92.83} | Helped |
| Use NTKSketch instead of RFF | kernel ridge regression · WorkLoads | MSE: $4.034\times 10^{4}$ -> $3.538\times 10^{4}$ | Helped |
| Use NTKSketch instead of NTKRF | kernel ridge regression · WorkLoads | MSE: $3.554 \times 10^{4}$ -> $3.538\times 10^{4}$ | Helped |
| Use NTK approximations instead of exact NTK | kernel ridge regression · CT | MSE: 30.52 -> 46.91; 30.52 -> 46.52 | Hurt |
| Use NTK approximations instead of exact NTK | kernel ridge regression · Protein | MSE: 20.24 -> 20.51; 20.24 -> 21.19 | Hurt |
| Use NTKRF instead of RFF | kernel ridge regression · CT | MSE: 48.20 -> 46.91 | Helped |
| Use NTKSketch instead of RFF | kernel ridge regression · CT | MSE: 48.20 -> 46.52 | Helped |
| Use NTK approximations instead of RFF | kernel ridge regression · Protein | MSE: 19.72 -> 20.51; 19.72 -> 21.19 | Mixed |
Closest papers by meaning
- Optimized Random Features for the Neural Tangent Kernel (Student Abstract) (AAAI 2025 · similarity 0.92)
- Harnessing the Power of Infinitely Wide Deep Nets on Small-data Tasks (ICLR 2020 · similarity 0.92)
- Fast Finite Width Neural Tangent Kernel (ICML 2022 · similarity 0.92)
- The Surprising Effectiveness of Infinite-Width NTKs for Characterizing and Improving Model Training (AAAI 2025 · similarity 0.89)
- On the Impacts of the Random Initialization in the Neural Tangent Kernel Theory (NeurIPS 2024 · similarity 0.89)
- Neural Tangent Kernels Motivate Cross-Covariance Graphs in Neural Networks (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.