{"slug": "kernel-reboot-breaking-the-boundaries-of-neural-tangent-kernels-for-neural", "title": "Kernel Reboot: Breaking the Boundaries of Neural Tangent Kernels for Neural Fields", "summary": "Researchers introduced three algorithms—NTK-KIP, MetaQuill, and MetaQuill-KIP—that extend Neural Tangent Kernel (NTK) regression to neural fields, enabling non-linear, meta-learnable reconstruction from sparse observations. MetaQuill-KIP, which fuses KIP-style warm starts with meta-learned initialization, achieves high-PSNR reconstructions and semantically plausible inpainting with lightweight per-instance adaptation, narrowing the gap between analytic kernels and practical few-shot reconstruction.", "body_md": "arXiv:2609.03117v1 Announce Type: new\nAbstract: Neural fields (NFs) map continuous coordinates to signals such as color or density, but fast high-quality reconstruction from sparse observations remains difficult. Classical Neural Tangent Kernel (NTK) regression gives closed-form fits, yet it is fundamentally linear and cannot accumulate reusable task priors. We develop three algorithms that address these gaps. NTK-KIP learns a distilled support set of coordinates (and optional labels) so that a finite NTK can inpaint large missing regions from little observed data, yielding a compact non-linear representation instead of a raw kernel solve. MetaQuill meta-learns a shared initialization for an INR so that new scenes can be adapted by updating only a small task-specific weight offset, which provides true feature learning and a reusable prior. Finally, MetaQuill-KIP fuses both ideas: it seeds the task with a KIP-style non-linear warm start, then refines only that small offset around the meta-learned initialization. MetaQuill-KIP achieves high-PSNR reconstructions and semantically plausible inpainting under very sparse observations, while requiring only lightweight per-instance adaptation, whereas diffusion-style baselines typically depend on large pretrained generative priors and costly per-image tuning. This shows that NTK-driven neural fields can be made both non-linear and meta-learnable, narrowing the gap between analytic kernels and practical few-shot reconstruction.", "url": "https://wpnews.pro/news/kernel-reboot-breaking-the-boundaries-of-neural-tangent-kernels-for-neural", "canonical_source": "https://arxiv.org/abs/2609.03117", "published_at": "2026-09-04 04:00:00+00:00", "updated_at": "2026-09-04 04:25:25.335381+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "ai-research"], "entities": ["arXiv", "NTK-KIP", "MetaQuill", "MetaQuill-KIP"], "alternates": {"html": "https://wpnews.pro/news/kernel-reboot-breaking-the-boundaries-of-neural-tangent-kernels-for-neural", "markdown": "https://wpnews.pro/news/kernel-reboot-breaking-the-boundaries-of-neural-tangent-kernels-for-neural.md", "text": "https://wpnews.pro/news/kernel-reboot-breaking-the-boundaries-of-neural-tangent-kernels-for-neural.txt", "jsonld": "https://wpnews.pro/news/kernel-reboot-breaking-the-boundaries-of-neural-tangent-kernels-for-neural.jsonld"}}