Lune

ICLR2025Top-tier venue

How Learnable Grids Recover Fine Detail in Low Dimensions: A Neural Tangent Kernel Analysis of Multigrid Parametric Encodings

Samuel Audia, Soheil Feizi, Matthias Zwicker, Dinesh Manocha

Abstract

Neural networks that map between low dimensional spaces are ubiquitous in computer graphics and scientific computing; however, in their naive implementation, they are unable to learn high frequency information. We present a comprehensive analysis comparing the two most common techniques for mitigating this spectral bias: Fourier feature encodings (FFE) and multigrid parametric encodings (MPE). FFEs are seen as the standard for low dimensional mappings, but MPEs often outperform them and learn representations with higher resolution and finer detail. FFE's roots in the Fourier transform, make it susceptible to aliasing if pushed too far, while MPEs, which use a learned grid structure, have no such limitation. To understand the difference in performance, we use the neural tangent kernel (NTK) to evaluate these encodings through the lens of an analogous kernel regression. By finding a lower bound on the smallest eigenvalue of the NTK, we prove that MPEs improve a network's performance through the structure of their grid and not their learnable embedding. This mechanism is fundamentally different from FFEs, which rely solely on their embedding space to improve performance. Results are empirically validated on a 2D image regression task using images taken from 100 synonym sets of ImageNet and 3D implicit surface regression on objects from the Stanford graphics dataset. Using peak signal-to-noise ratio (PSNR) and multiscale structural similarity (MS-SSIM) to evaluate how well fine details are learned, we show that the MPE increases the minimum eigenvalue by 8 orders of magnitude over the baseline and 2 orders of magnitude over the FFE. The increase in spectrum corresponds to a 15 dB (PSNR) / 0.65 (MS-SSIM) increase over baseline and a 12 dB (PSNR) / 0.33 (MS-SSIM) increase over the FFE.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

1 top-tier paper has cited it since. Know what they found before your reviewers do.

Questions to start from

Lune can draw on

the full paper · the top-tier papers that cite it · related work

Get the answer

Free to start. No credit card required.

Lineage

  • Mip-NeRF: A Multiscale Representation for Anti-Aliasing Neural Radiance Fields cites Fourier Features Let Networks Learn High Frequency Functions in Low Dimensional Domains
  • Instant neural graphics primitives with a multiresolution hash encoding cites Real-time neural radiance caching for path tracing
  • Instant neural graphics primitives with a multiresolution hash encoding cites Mip-NeRF: A Multiscale Representation for Anti-Aliasing Neural Radiance Fields
  • Instant neural graphics primitives with a multiresolution hash encoding cites Fourier Features Let Networks Learn High Frequency Functions in Low Dimensional Domains
  • Unveiling the Mechanism of Continuous Representation Full-Waveform Inversion: A Wave Based Neural Tangent Kernel Framework cites Instant neural graphics primitives with a multiresolution hash encoding
  • Plenoxels: Radiance Fields without Neural Networks cites Mip-NeRF: A Multiscale Representation for Anti-Aliasing Neural Radiance Fields
Builds onCited byLeft to right by year; the more cited, the higher.

Cited by top-tier papers1

  1. Unveiling the Mechanism of Continuous Representation Full-Waveform Inversion: A Wave Based Neural Tangent Kernel FrameworkRuihua Chen, Yisi Luo, Bangyu Wu, Deyu MengICLR 2026 · 2 citations

Builds on9

  1. Instant neural graphics primitives with a multiresolution hash encodingThomas Müller, Alex Evans, Christoph Schied, Alexander KellerSIGGRAPH 2022 · 4,089 citations
  2. Fourier Features Let Networks Learn High Frequency Functions in Low Dimensional DomainsMatthew Tancik, Pratul P. Srinivasan, Ben Mildenhall, Sara Fridovich-Keil et al.NeurIPS 2020 · 4,036 citations
  3. Mip-NeRF: A Multiscale Representation for Anti-Aliasing Neural Radiance FieldsJonathan T. Barron, Ben Mildenhall, Matthew Tancik, Peter Hedman et al.ICCV 2021 · 2,700 citations
  4. Plenoxels: Radiance Fields without Neural NetworksSara Fridovich-Keil, Alex Yu, Matthew Tancik, Qinhong Chen et al.CVPR 2022 · 1,237 citations
  5. Real-time neural radiance caching for path tracingThomas Müller, Fabrice Rousselle, Jan Novák, Alexander KellerSIGGRAPH 2021 · 140 citations

Related papers

  1. The Spectral Bias of Polynomial Neural NetworksMoulik Choraria, Leello Tadesse Dadi, Grigorios Chrysos, Julien Mairal et al.ICLR 2022 · 26 citations
  2. Extrapolation and Spectral Bias of Neural Nets with Hadamard Product: a Polynomial Net StudyYongtao Wu, Zhenyu Zhu, Fanghui Liu, Grigorios Chrysos et al.NeurIPS 2022 · 19 citations
  3. Learning High-Frequency Functions Made Easy with Sinusoidal Positional EncodingChuanhao Sun, Zhihang Yuan, Kai Xu, Luo Mai et al.ICML 2024 · 12 citations
  4. Separable Neural Networks: Approximation Theory, NTK Regime, and Preconditioned Gradient DescentYisi Luo, Deyu MengICLR 2026
  5. A Structured Dictionary Perspective on Implicit Neural RepresentationsGizem Yüce, Guillermo Ortiz-Jiménez, Beril Besbinar, Pascal FrossardCVPR 2022 · 62 citations
Dusk over the sea between two cliffs drawn in fine vertical lines