Pith. sign in

REVIEW 3 major objections 5 minor 39 references

Multi-resolution Enhancement for Full Spectrum Neural Representations

T0 review · 3 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read By splitting signals into wavelet scales and adding a kernel-prediction network at the finest band, WIEN-INR lets compact neural representations keep fine speckle detail that standard INRs blur, with higher PSNR at matched compression.

desk verdict A genuinely useful multi-scale INR architecture for scientific data, with a real external-validity caveat: the enhancement module assumes cross-scale wavelet correlation, so 'data-agnostic' overclaims. read the letter →

arxiv 2509.15494 v2 pith:I6JI7FR5 submitted 2025-09-19 cs.LG physics.data-an

Yuan Ni , Zhantao Chen , Shizhou Xu , Cheng Peng , Rajan Plumley , Chun Hong Yoon , Jana B. Thayer , Joshua J. Turner This is my paper
classification cs.LGphysics.data-an
keywords implicitneuralrepresentationwavelettransformmulti-resolutionanalysisscientificdatacompressionrate-distortionX-rayscatteringneutronenhancementnetwork
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

WIEN-INR is an attempt to get compact implicit neural representations to reproduce the full frequency content of scientific measurements, including sharp speckles and textures that small networks usually over-smooth. It applies a multi-level wavelet transform to split the data into frequency bands, trains a small coordinate network per band, and adds a lightweight enhancement module that predicts the finest detail band from an upsampled coarser prediction using a learned local kernel. On three experimental datasets—wide-angle X-ray diffraction, inelastic neutron scattering, and ultra-fast X-ray scattering—the authors report higher PSNR and SSIM at matched compression ratios than four standard INR baselines, along with preservation of the wavelet coefficient distribution. If correct, the method offers a practical way to store and query large scientific volumes as network weights without losing the fine structure that carries physical information.

What carries the argument

Enhancement module (learned local deconvolution). For each orientation and scale transition, a lightweight SIREN takes a coordinate x and emits an r^p kernel; the kernel is applied as an inner product to an r×r patch extracted from the upsampled prediction of the next-coarser INR, and the result is trained to match the true wavelet coefficient at x. This is the mechanism that carries the claim of recovering subtle detail without inflating model size; it exploits the geometric regularity of wavelet coefficients—large amplitudes cluster along contours and textures, so fine detail is locally predictable from coarse structure. The multi-resolution preprocess (DWT plus one shared INR per scale fo

What would settle it

Take a real dataset, apply the wavelet transform, and randomly permute the spatial positions of the finest detail coefficients while keeping their marginal statistics intact; train WIEN-INR and the same framework without enhancement. If the enhancement no longer improves PSNR (or worsens it), the local coarse-to-fine correlation assumption is exposed as the source of the gain. A cleaner version: generate a volume whose finest-scale wavelet coefficients are independent random noise—there, the module has nothing predictable to learn and should show no rate-distortion benefit.

Watch

Extended reading notes

Core claim

WIEN-INR claims that the low-frequency bias of compact INRs can be overcome not by enlarging the network, but by reorganizing the representation: a discrete wavelet transform separates the signal into a coarse approximation and orientation-specific detail bands; each band is assigned its own small SIREN, with frequency parameters matched to the band; and the finest band is synthesized by an enhancement network rather than learned directly. The enhancement up-samples the next-coarser scale's prediction and applies a per-location r×r kernel—outputs of a separate lightweight network—to the local patch, matching the true finest coefficients. Across three raw experimental datasets, this yields hi

Load-bearing premise

The load-bearing premise is that fine-scale wavelet detail is locally predictable from the next-coarser scale through a small learned kernel; if a dataset's finest structures are spatially independent of coarser structure, the enhancement module cannot recover them and the central fidelity gain disappears.

Editorial extensions

If this is right

  • Raw experimental volumes can be replaced by a set of small network weights, with decoding as a forward pass, and at a fixed weight budget the reconstructed data retains more fine-scale structure than with a single large INR.
  • Because the objective separates by scale, subnetworks can be trained in parallel; only the enhancement step depends on the previous scale, so encoding time can drop relative to monolithic INR training.
  • Each scale is a continuous coordinate function, so users can decode arbitrary regions of interest or resolutions without retraining, and the finest band is enhanced rather than emulated from scratch.
  • The enhancement module is modular, so existing coordinate-based INR pipelines can adopt it without changing their core architecture.
  • Preserving the wavelet-coefficient distribution, not just pixel error, means compression preserves multi-scale statistics that downstream physical analyses rely on.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the local coarse-to-fine correlation the enhancement assumes holds broadly, WIEN-INR effectively learns a data-driven zerotree-style prior for INR compression; combining it with entropy-coded weight quantization could push scientific INR codecs close to wavelet transform-codec rate-distortion limits.
  • The same coarse-to-fine predictor suggests a super-resolution mode the paper does not test: train only coarse scales, then let the enhancement module synthesize finer detail at inference, yielding resolution-scalable scientific data.
  • On data whose finest band is dominated by detector noise rather than geometric regularity, the enhancement could fabricate texture rather than recover signal; a synthetic white-noise-detail experiment would delineate the regime in which WIEN-INR should be used.
  • The reported robustness across wavelet bases masks a tunable resource-allocation question: how to split the parameter budget across scales. An adaptive rule based on per-band energy could extend the method to datasets with very different spectra.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes WIEN-INR, a hierarchical implicit neural representation for scientific data compression. A discrete wavelet transform partitions the data into approximation and detail sub-bands; each band is modeled by a compact SIREN, with the finest detail band predicted by an enhancement module that applies a coordinate-dependent kernel to the upsampled output of the next-coarser scale. Experiments on three scattering datasets (X-ray diffraction, neutron scattering, ultrafast X-ray scattering) report higher PSNR at matched compression ratios than FE, SIREN, WIRE, and TCNN, together with robustness to wavelet basis and decomposition depth.

Significance. If the reported results are reproducible, WIEN-INR offers a practical improvement to INR-based compression for scientific data. The modular design (wavelet decomposition as preprocessing, optional enhancement module) is clean and the ablation against Wavelet-INR in Fig. 6 is valuable evidence that the enhancement module adds representational capacity beyond the multi-scale decomposition itself. The paper also includes a useful robustness analysis (Fig. 7) for wavelet type, number of scales, and window size. No code is provided and no theoretical guarantees are claimed, so the contribution is empirical.

major comments (3)
  1. [Sec. III.a, Eqs. (2) and (3)] The sums run over j=1..J−1. With the indexing used in the paper (j=1 finest, j=J coarsest; see Sec. III.c where the enhancement module predicts d1 from d2), a J-level DWT has J detail sub-bands d1..dJ. As written, the coarsest detail band d_J is omitted from the training objective. This is load-bearing: it changes which coefficients the networks are trained to represent. Please correct to j=1..J, or precisely define a different indexing convention and apply it consistently.
  2. [Sec. IV.b, Fig. 4] The central rate-distortion claim rests on single-run curves. No number of seeds, error bars, or statistical significance tests are reported. Since the differences between methods at a given compression ratio appear modest and the baseline precision is not specified, please provide multiple independent runs and report mean ± std, or at least state the seed variability.
  3. [Sec. VI vs Sec. IV.a] The conclusion claims 'data-agnostic applicability,' but the enhancement module (Sec. III.c, Eq. 5) assumes fine-scale wavelet coefficients are locally predictable from coarser-scale coefficients (a zerotree-like prior). The paper itself states in Sec. IV.a that applying the enhancement to the finest band is a data-dependent choice. All three evaluation datasets are scattering measurements with speckle/texture, which plausibly share this cross-scale structure. To support the 'data-agnostic' claim, either test on data without such parent-child correlation (e.g., i.i.d. noise, natural images) or temper the conclusion to specify the applicable data regime.
minor comments (5)
  1. [Sec. III.c, Eq. (5)] The notation 'N∇(x)' should be 'N_r(x)' for the r×r neighborhood. Also clarify whether the patch is extracted from φ^i_j or its upsampled version φ^i_{j−1}; the current text and equation are inconsistent.
  2. [Fig. 4a caption] The caption says 'coarse levels (a and d1)', but elsewhere d1 is the finest detail band (e.g., Sec. III.c describes enhancing 'd2 to d1'). Please clarify which scale is meant, or correct the indexing.
  3. [Table I] The enhancement window size is given as '(2^1+1)^p to (2^3+1)^p'; this is a roundabout way to say 3^p to 9^p. Please simplify for readability.
  4. [Sec. V] The description 'post-training quantization of network weights (division by two)' is unclear. Division by two is a scaling operation, not quantization to half precision. Please specify the actual quantization/rounding scheme used.
  5. [References] Reference [30] appears to duplicate reference [7] (both cite the same paper by Lu et al. on compressive neural representations of volumetric scalar fields). Please consolidate.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: WIEN-INR's enhancement prediction is a learned conditional model (Eq. 5) benchmarked against external baselines, so no claimed result reduces to its inputs by construction; one non-load-bearing self-citation and a data-dependent architecture choice are the only mild residues.

full rationale

Walking the derivation chain: the wavelet decomposition (Eqs. 2-3) maps the data into coefficient sub-bands; the per-scale SIREN subnetworks regress those coefficients; and the enhancement module (Eqs. 4-5) predicts the finest detail band d^i_{j-1}(x) as an inner product of an upsampled coarser-scale prediction patch with a learned kernel output by P^i_{j->j-1,xi}(x), trained by minimizing the squared error against the true coefficients. Nothing here is defined in terms of its output: the kernel network is genuinely fit, and the reported fidelity (PSNR/SSIM at matched compression ratio, Fig. 4b) is measured against external baselines (FE, SIREN, WIRE, TCNN) in a domain-neutral pixel space, so the central claim is not a fitted input renamed as a prediction. The zerotree/geometric-regularity prior motivating the module is imported from standard external references (Mallat [16]; Shapiro [17]; JPEG-2000 [23]), not from the authors' prior work, so no ansatz is smuggled in via self-citation. The only self-citation is [25], used analogically ('reminiscent of a progressive refinement') in Sec. III.c; it is not load-bearing, so per the rubric it only nudges the score to 2. Two non-circular qualifications are flagged per the reviewing rule: (i) Sec. IV.a concedes the key design choice is data-dependent ('We choose to apply the enhancement structure to the finest detail band... This choice is data-dependent... but we found it sufficient for all evaluated scientific datasets'), which undercuts the Sec. VI claim of 'data-agnostic applicability' and is an external-validity risk for data lacking cross-scale wavelet correlation; (ii) Fig. 4a's wavelet-coefficient-fidelity plot measures WIEN-INR in the same domain in which it was trained, so it partly restates the training objective, though the headline rate-distortion claims are made in a domain-neutral pixel space. No equation or uniqueness theorem in the paper reduces a claimed prediction to a fitted parameter, and no self-citation chain forces the architecture choice. Verdict: no significant circularity.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The method's success depends on standard wavelet analysis (invertibility, multiresolution), the empirical spectral-bias behavior of coordinate MLPs, and the geometric-regularity assumption that fine wavelet coefficients are locally predictable from coarser scales. The free parameters (J, n_r, wavelet basis, per-scale widths, SIREN frequencies) are hand-tuned rather than derived. No new physical entities are introduced.

free parameters (5)
  • Number of wavelet scales J = 4
    Chosen as best balance of frequency separation; tested J in {2,3,4} on Cu3Au data.
  • Enhancement window size n_r = 3
    Neighborhood radius for kernel prediction; tested n_r in {3,5,9}.
  • Wavelet basis = Haar
    Haar at J=4, n_r=3 yields strong results across benchmark datasets; five bases compared on one dataset.
  • Per-scale hidden layer widths = 24 to 1024
    Parameter budget allocated across scales: coarser scales get smaller subnetworks, finer scales larger; exact allocation per dataset not specified.
  • SIREN frequency parameters = omega0=30 for enhancement net; per-scale frequencies varied
    Hyperparameters controlling frequency band alignment; chosen by hand, not derived.
assumptions (5)
  • standard math DWT with a chosen basis is invertible and provides a hierarchical time-frequency decomposition.
    Relied on in Sec. II.b and III.a; the whole encoding/decoding pipeline depends on being able to recover the signal from predicted coefficients.
  • standard math Coordinate MLPs with sufficient width can approximate any measurable function (universal approximation).
    Invoked in Sec. I to motivate why large INRs could in principle represent full spectra, and to justify using MLPs for each sub-band.
  • standard math Optimization of coordinate networks exhibits spectral bias, fitting low frequencies first.
    Invoked in Sec. II.c citing [1,3]; this motivates the wavelet domain separation.
  • domain assumption Signals of interest (scientific images) obey geometric regularity: significant wavelet coefficients cluster along contours and are locally correlated across scales.
    Invoked in Sec. III.b and III.c as 'geometric image regularity' and 'zerotree coding' to justify using the coarser scale as a prior for the finer scale.
  • domain assumption The fine detail band [pi/2, pi] is the most difficult to represent, and the difficulty is representational capacity rather than frequency bias.
    Stated in Sec. III.c 'From our experiments, we observe...'; motivates applying enhancement only at the finest band and is not proven.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multi-resolution Enhancement for Full Spectrum Neural Representations." pith.science (2026). https://pith.science/paper/I6JI7FR5

@misc{pith2026250915494,
  author       = {Pith},
  title        = {Pith review of: Multi-resolution Enhancement for Full Spectrum Neural Representations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I6JI7FR5}},
  note         = {Machine review of arXiv:2509.15494}
}
read the original abstract

Scientific data acquisition continues to outpace storage and analysis capabilities, making voxel-based representations increasingly intractable. Implicit neural representations (INRs) offer a promising solution by encoding signals through coordinate-based neural networks, serving as surrogates of data, with computational and storage requirements scaling with network complexity rather than data dimensionality. However, smaller INRs struggle to faithfully represent the multi-scale structures, high-frequency information, and fine textures that constitute a large proportion of scientific measurements. We propose WIEN-INR, a theoretically-guided hierarchical INR framework that distributes modeling across resolution scales and enables improved representation capacity through a novel enhancement network to recover subtle details. This multi-scale architecture allows smaller networks to retain the full spectrum of information while preserving training efficiency and lowering storage cost. Evaluated on distinct raw experimental measurements across scales and complexities, WIEN-INR represents a practical step toward broader adoption of neural representations in scientific workflows, delivering compact, robust, and high-fidelity representations.

Figures

Figures reproduced from arXiv: 2509.15494 by the authors.

Figure 1
Figure 1. FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: presents the main result, showing our method consistently achieves superior rate–distortion performance compared to all benchmarks. Additionally, we compare pixel-wise differences and correlation with ground truth ( [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6. Results are shown for the neutron scattering [27](left [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7. Robustness to wavelet choices. (Left) Rate–distortion [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

39 extracted references · 1 canonical work pages

  1. [1]

    Fourier features let networks learn high frequency func- tions in low dimensional domains

    Matthew Tancik, Pratul P Srinivasan, Ben Mildenhall, Sara Fridovich-Keil, Nithin Raghavan, Utkarsh Singhal, Ravi Ramamoorthi, Jonathan T Barron, and Ren Ng. Fourier features let networks learn high frequency func- tions in low dimensional domains. Advances in Neural Information Processing Systems, 2020

    2020

  2. [2]

    Baraniuk

    Vishwanath Saragadam, Daniel LeJeune, Jasper Tan, Guha Balakrishnan, Ashok Veeraraghavan, and Richard G. Baraniuk. WIRE: Wavelet implicit neural representations. In 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) ,

    2023

  3. [3]

    Implicit neural representations with periodic activation functions

    Vincent Sitzmann, Julien Martel, Alexander Bergman, David Lindell, and Gordon Wetzstein. Implicit neural representations with periodic activation functions. In Advances in Neural Information Processing Systems, vol- ume 33, pages 7462–7473. Curran Associates, Inc., 2020

    2020

  4. [4]

    Dami- ani, Christopher Ford, M

    Jana Thayer, Zhantao Chen, Richard Claus, D. Dami- ani, Christopher Ford, M. Dubrovin, Victor Elmir, W. Kroeger, Xiang Li, S. Marchesini, Valerio Mariani, Riccardo Melcchiori, Silke Nelson, Ariana Peck, A. Per- azzo, F. Poitevin, Christopher P. O’Grady, Julieth Otero, Omar Quijano, Murali Shankar, M. Uervirojnangkoorn, Riccardo Veraldi, Matthew Weaver, C....

    arXiv 2024

  5. [5]

    Data reduction activities at European XFEL: Early results

    Egor Sobolev, Philipp Schmidt, Janusz Malka, David Hammer, Djelloul Boukhelef, Johannes M¨ oller, Karim Ahmed, Richard Bean, Ivette Jazm ´ ın Berm´ udez Mac ´ ıas, Johan Bielecki, and et al. Data reduction activities at European XFEL: Early results. Frontiers in Physics, 12, Feb 2024. doi:10.3389/fphy.2024.1331329

    arXiv 2024

  6. [6]

    Where do we stand with implicit neural representations? a technical and performance survey, 2025

    Amer Essakine, Yanqi Cheng, Chun-Wun Cheng, Lipei Zhang, Zhongying Deng, Lei Zhu, Carola-Bibiane Sch¨ onlieb, and Angelica I Aviles-Rivero. Where do we stand with implicit neural representations? a technical and performance survey, 2025. URL https://arxiv. org/abs/2411.03688

    arXiv 2025

  7. [8]

    Implicit neural representa- tions for image compression

    Yannick Str¨ umpler, Janis Postels, Ren Yang, Luc Van Gool, and Federico Tombari. Implicit neural representa- tions for image compression. In Shai Avidan, Gabriel Brostow, Moustapha Ciss´ e, Giovanni Maria Farinella, and Tal Hassner, editors, Computer Vision – ECCV 2022, pages 74–91, Cham, 2022. Springer Nature Switzer- land. ISBN 978-3-031-19809-0

    2022

  8. [9]

    Instant neural graphics primitives with a multiresolution hash encoding

    Thomas M¨ uller, Alex Evans, Christoph Schied, and Alexander Keller. Instant neural graphics primitives with a multiresolution hash encoding. ACM Transac- tions on Graphics (TOG), 41:1 – 15, 2022. URL https: //api.semanticscholar.org/CorpusID:246016186

    2022

Show all 39 references
  1. [10]

    Srinivasan, Matthew Tan- cik, Jonathan T

    Ben Mildenhall, Pratul P. Srinivasan, Matthew Tan- cik, Jonathan T. Barron, Ravi Ramamoorthi, and Ren Ng. NeRF: Representing scenes as neural radiance fields for view synthesis. In Andrea Vedaldi, Horst Bischof, Thomas Brox, and Jan-Michael Frahm, editors, Com- puter Vision – ...

    2020

  2. [11]

    anomalies,

    : the minimum number of bits (rate) required to represent data depends on how much information can be altered or discarded without exceeding a specified error threshold. Such rates are in turn determined by network size, weight quantization, and entropy coding [8, 12]. In the ...

    2025 arXiv

  3. [12]

    Entropy-constrained implicit neural representations for deep image compression

    Soonbin Lee, Jong-Beom Jeong, and Eun-Seok Ryu. Entropy-constrained implicit neural representations for deep image compression. IEEE Signal Processing Let- ters, 30:663–667, 2023. doi:10.1109/LSP.2023.3279780

    2023

  4. [13]

    George V. Cybenko. Approximation by superpositions of a sigmoidal function. Mathematics of Control, Sig- 11 nals and Systems, 2:303–314, 1989. URL https://api. semanticscholar.org/CorpusID:3958369

    1989

  5. [14]

    Cover and Joy A

    Thomas M. Cover and Joy A. Thomas. Elements of In- formation Theory. Wiley-Interscience, 2 edition, 2006. doi:https://doi.org/10.1002/047174882X.ch10

    2006 doi

  6. [15]

    Lossy compression of scientific data: Applications constrains and requirements, 2025

    Franck Cappello, Allison Baker, Ebru Bozda, Martin Burtscher, Kyle Chard, Sheng Di, Paul Christopher O Grady, Peng Jiang, Shaomeng Li, Erik Lindahl, Peter Lindstrom, Magnus Lundborg, Kai Zhao, Xin Liang, Masaru Nagaso, Kento Sato, Amarjit Singh, Seung Woo Son, Dingwen Tao, Jia...

    2025 arXiv

  7. [16]

    A Wavelet Tour of Signal Processing, Third Edition: The Sparse Way

    Stphane Mallat. A Wavelet Tour of Signal Processing, Third Edition: The Sparse Way. Academic Press, Inc., USA, 3rd edition, 2008. ISBN 0123743702

    2008

  8. [17]

    HOSC: A periodic activation function for pre- serving sharp features in implicit neural representations,

    Danzel Serrano, Jakub Szymkowiak, and Przemyslaw Musialski. HOSC: A periodic activation function for pre- serving sharp features in implicit neural representations,

  9. [18]

    FINER: Flexible spectral-bias tuning in implicit neural representation by variable-periodic activation functions

    Zhen Liu, Hao Zhu, Qi Zhang, Jingde Fu, Weibing Deng, Zhan Ma, Yanwen Guo, and Xun Cao. FINER: Flexible spectral-bias tuning in implicit neural representation by variable-periodic activation functions. In Proceedings of the IEEE/CVF Computer Vision and Pattern Recogni- tion Co...

    2024

  10. [19]

    InCode: Implicit neural conditioning with prior knowledge embeddings

    Amirhossein Kazerouni, Reza Azad, Alireza Hosseini, Dorit Merhof, and Ulas Bagci. InCode: Implicit neural conditioning with prior knowledge embeddings. In Pro- ceedings of the IEEE/CVF Winter Conference on Appli- cations of Computer Vision (WACV), pages 1298–1307, 2024

    2024

  11. [20]

    Improved implicit neural representation with fourier reparameter- ized training

    Kexuan Shi, Xingyu Zhou, and Shuhang Gu. Improved implicit neural representation with fourier reparameter- ized training. In Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition, pages 25985–25994, 2024

    2024

  12. [21]

    J.M. Shapiro. Embedded image coding using zerotrees of wavelet coefficients. IEEE Transactions on Signal Pro- cessing, 41(12):3445–3462, 1993. doi:10.1109/78.258085

    1993 doi

  13. [22]

    Cross-frequency implicit neural representation with self- evolving parameters, 2025

    Chang Yu, Yisi Luo, Kai Ye, Xile Zhao, and Deyu Meng. Cross-frequency implicit neural representation with self- evolving parameters, 2025. URL https://arxiv.org/ abs/2504.10929

    2025 arXiv

  14. [23]

    Taubman and Michael W

    David S. Taubman and Michael W. Marcellin. JPEG2000: Image Compression Fundamentals, stan- dards, and Practice. Springer Science + Business Media, LLC, 2013

    2013

  15. [24]

    Coifman and M.V

    R.R. Coifman and M.V. Wickerhauser. Entropy-based algorithms for best basis selection. IEEE Transac- tions on Information Theory, 38(2):713–718, 1992. doi: 10.1109/18.119732

    1992 doi

  16. [25]

    Multiplicative filter networks

    Rizal Fathony, Anit Kumar Sahu, Devin Willmott, and J Zico Kolter. Multiplicative filter networks. In Inter- national Conference on Learning Representations, 2021. URL https://openreview.net/forum?id=OmtmcPkkhT

    2021

  17. [26]

    Testing the data framework for an AI algorithm in preparation for high data rate X- ray facilities

    Hongwei Chen et al. Testing the data framework for an AI algorithm in preparation for high data rate X- ray facilities. In 4th Annual Workshop on Extreme- scale Experiment-in-the-Loop Computing, 10 2022. doi: 10.1109/XLOOP56614.2022.00006

    2022

  18. [27]

    High-energy Spin Waves in the Spin-1 Square- lattice Antiferromagnet La 2NiO4

    AN Petsch, NS Headings, D Prabhakaran, AI Kolesnikov, CD Frost, AT Boothroyd, R Coldea, and SM Hay- den. High-energy Spin Waves in the Spin-1 Square- lattice Antiferromagnet La 2NiO4. Physical Review Re- search, 5(3):033113, 2023. URL https://doi.org/10. 1103/PhysRevResearch.5.033113

    2023

  19. [28]

    Song Han, Huizi Mao, and William J. Dally. Deep compression: Compressing deep neural network with pruning, trained quantization and huffman coding. International Conference on Learning Representations (ICLR), 2016. URL https://api.semanticscholar. org/CorpusID:2134321

    2016

  20. [29]

    Petsch, Ed- mund Xu, Cheng Peng, Alexander I

    Yuan Ni, Zhantao Chen, Alexander N. Petsch, Ed- mund Xu, Cheng Peng, Alexander I. Kolesnikov, Sugata Chowdhury, Arun Bansil, Jana B. Thayer, and Joshua J. Turner. Physics-guided dual implicit neural representa- tions for source separation, 2025. URL https://arxiv. org/abs/2507.05249

    2025 arXiv

  21. [30]

    Y. Lu, K. Jiang, J. A. Levine, and M. Berger. Com- pressive neural representations of volumetric scalar fields. Computer Graphics Forum, 40(3), 06 2021. ISSN ISSN 0167-7055. doi:10.1111/cgf.14295

    2021 doi

  22. [31]

    Sparsity in deep learning: pruning and growth for efficient inference and training in neural networks

    Torsten Hoefler, Dan Alistarh, Tal Ben-Nun, Nikoli Dry- den, and Alexandra Peste. Sparsity in deep learning: pruning and growth for efficient inference and training in neural networks. Journal of Machine Learning Research, 2021

    2021

  23. [32]

    DCINR: a divide-and-conquer implicit neural representation for compressing time- varying volumetric data in hours

    Jun Han and Fan Yang. DCINR: a divide-and-conquer implicit neural representation for compressing time- varying volumetric data in hours. IEEE transactions on visualization and computer graphics, PP, 2025. doi: 10.1109/TVCG.2025.3564255

    2025

  24. [33]

    Deep im- age compression in the wavelet transform domain based on high frequency sub-band prediction

    Chuxi Yang, Yan Zhao, and Shigang Wang. Deep im- age compression in the wavelet transform domain based on high frequency sub-band prediction. IEEE Access, 7: 52484–52497, 2019. doi:10.1109/ACCESS.2019.2911403

    2019

  25. [34]

    WINNet: Wavelet-inspired invertible network for image denoising

    Jun-Jie Huang and Pier Luigi Dragotti. WINNet: Wavelet-inspired invertible network for image denoising. Trans. Img. Proc., 31, 2022. ISSN 1057-7149. doi: 10.1109/TIP.2022.3184845. URL https://doi.org/10. 1109/TIP.2022.3184845

    2022

  26. [35]

    Hou, and Max Tegmark

    Ziming Liu, Yixuan Wang, Sachin Vaidya, Fabian Ruehle, James Halverson, Marin Soljaˇ ci´ c, Thomas Y. Hou, and Max Tegmark. KAN: Kolmogorov–arnold net- works. In International Conference on Learning Rep- resentations (ICLR), 2025. URL https://openreview. net/forum?id=Ozo7qJ5vZi. Oral

    2025

  27. [36]

    Neural wavelet-domain diffusion for 3D shape generation

    Ka-Hei Hui, Ruihui Li, Jingyu Hu, and Chi-Wing Fu. Neural wavelet-domain diffusion for 3D shape generation. December 2022

    2022

  28. [37]

    Suo, and Qionghai Dai

    Runzhao Yang, Tingxiong Xiao, Yuxiao Cheng, J. Suo, and Qionghai Dai. TINC: Tree-structured implicit neural compression. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 18517– 18526, 2022. doi:10.1109/CVPR52729.2023.01776

    2023

  29. [41]

    Generative neural fields by mixtures of neu- ral implicit functions, 2023

    Tackgeun You, Mijeong Kim, Jungtaek Kim, and Bo- hyung Han. Generative neural fields by mixtures of neu- ral implicit functions, 2023. URL https://arxiv.org/ abs/2310.19464

    2023 arXiv

  30. [2020]

    ISBN 978-3-030- 58452-8

    Springer International Publishing. ISBN 978-3-030- 58452-8

  31. [2024]

    URL https://arxiv.org/abs/2401.10967

    arXiv

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.