Research map / Learning theory and algorithms / Kernels and regression theory
Kernel Reboot: Breaking the Boundaries of Neural Tangent Kernels for Neural Fields
Amir Mallak, Alaa Maalouf, Lior Wolf, Daniela L. Rus, Dan Rosenbaum
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 |
|---|---|---|---|
| add nonlinear refinement of delta-theta after KIP initialization (ablation) | MetaQuill-KIP · MNIST held-out digit | PSNR (dB): 11.31 -> 32.49 | Improved |
| initialize delta-theta from KIP support instead of KRR solution (ablation) | MetaQuill-KIP · MNIST held-out digit | PSNR (dB): 26.65 -> 32.49 | Improved |
| add nonlinear refinement beyond tangent-only meta adaptation (ablation) | MetaQuill-KIP · MNIST held-out digit | PSNR (dB): 18.32 -> 32.49 | Improved |
| add KIP-initialized nonlinear refinement over Single-INR | MetaQuill-KIP · Flowers full reconstruction | PSNR (dB): 13.57 -> 26.29 | Improved |
| add nonlinear refinement to KIP init (ablation) | MetaQuill-KIP · Flowers full reconstruction | PSNR (dB): 9.37 -> 26.29 | Improved |
| add nonlinear refinement to tangent-only adaptation (ablation) | MetaQuill-KIP · Flowers full reconstruction | PSNR (dB): 15.06 -> 26.29 | Improved |
| MetaQuill-KIP vs Single-INR(masked) at high observed fraction | MetaQuill-KIP · Flowers inpainting (observed fraction 0.80) | PSNR (dB): 14.00 -> 21.71 | Improved |
| MetaQuill-KIP vs Single-INR(masked) at very sparse observation | MetaQuill-KIP · Flowers inpainting (observed fraction 0.10) | PSNR (dB): 14.74 -> 11.97 | Hurt |
| MetaQuill-KIP vs Single-INR(masked) at 1% observed | MetaQuill-KIP · Flowers inpainting (observed fraction 0.01) | PSNR (dB): 13.60 -> 9.63 | Hurt |
| replace NTK regression with trained NF on circular-hole inpainting | Neural field · Flowers circular hole | PSNR (dB): 17.06 vs 20.21 | Mixed |
| use FPE instead of raw PE for NTK-KIP at high distillation (ablation) | NTK-KIP · Flowers distillation 95% | PSNR (dB): 19.81 -> 20.43 | Improved |
| use FPE instead of raw PE for NTK-KIP at 20% distillation (ablation) | NTK-KIP · Flowers distillation 20% | PSNR (dB): 33.35 -> 54.67 | Improved |
Closest papers by meaning
- How Learnable Grids Recover Fine Detail in Low Dimensions: A Neural Tangent Kernel Analysis of Multigrid Parametric Encodings (ICLR 2025 · similarity 0.84)
- Meta-Continual Learning of Neural Fields (ICLR 2025 · similarity 0.84)
- Learning Input Encodings for Kernel-Optimal Implicit Neural Representations (ICML 2025 · similarity 0.84)
- Decompositional Neural Scene Reconstruction with Generative Diffusion Prior (CVPR 2025 · similarity 0.84)
- Globally Convergent Variational Inference (NeurIPS 2024 · similarity 0.84)
- 3D Reconstruction with Generalizable Neural Fields using Scene Priors (ICLR 2024 · similarity 0.84)
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.
