Pith. sign in

REVIEW 3 major objections 5 minor 40 references

JFRFFNet: A Data-Model Co-Driven Graph Signal Denoising Model with Partial Prior Information

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 ↗ Review JSON

Pith's one-line read JFRFFNet embeds a Wiener filter into a neural network and learns the transform order pair and filter coefficients by backpropagation, enabling denoising of time-varying graph signals with only partial prior information.

desk verdict A coherent but incremental denoising architecture whose performance claims are undermined by post-hoc graph-matrix selection and a loose 'partial prior information' framing. read the letter →

arxiv 2509.09147 v1 pith:JC53ZLDO submitted 2025-09-11 eess.SP

Ziqi Yan , Zhichao Zhang This is my paper · ORCID
classification eess.SP
keywords graphsignaldenoisingjointtime-vertexfractionalFouriertransformWienerfilterlearnableorderneuralnetworkspatio-temporalprocessingbackpropagationpartialpriorinformation
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

The paper tries to establish that a classical model-based denoiser for time-varying graph signals—Wiener filtering in the joint time-vertex fractional Fourier (JFRFT) domain—can be turned into a trainable neural layer whose transform-order pair and filter coefficients are learned from data. If true, this would let practitioners denoise graph signals without knowing the signal and noise statistics in advance, needing only a modest set of clean/noisy training pairs. The central claim is that this data–model co-driven approach beats both the purely model-based Wiener filter and a graph neural network baseline across eight real-world datasets, ranking first or second in every case.

What carries the argument

The JFRFT matrix F^{α,β}_J = F^β ⊗ F^α_G, the Kronecker product of the discrete fractional Fourier matrix and the graph fractional Fourier matrix, serves as the core trainable transform. The paper derives exact derivatives of this matrix with respect to the fractional orders α and β, making the transform a fully differentiable layer. The Wiener filter is then approximated by a learnable pointwise (Hadamard) filter matrix H in the JFRFT domain, and both the orders and H are updated by backpropagation through forward and inverse transforms.

What would settle it

Train JFRFFNet on a dataset where the training set contains only noisy signals (no clean references) and measure the output SNR; if it fails to improve over the input SNR, the central claim of requiring only partial prior information is refuted. Additionally, the paper itself discards adjacency-based results on the METR dataset due to ill-conditioned eigendecomposition, so a dataset with such a matrix would break the method.

Watch

Extended reading notes

Core claim

The central discovery is that the JFRFT-domain Wiener filter can be embedded as a differentiable layer, with the two transform orders (α, β) and the diagonal filter matrix as learnable parameters. Because the JFRFT matrix is defined through matrix exponentials of graph and temporal fractional Fourier operators, the paper derives exact derivatives of the layer with respect to α and β and uses them in gradient descent. This turns a filter that previously required complete prior knowledge of signal and noise covariances into a network that learns the right transform domain and filter shape from paired clean/noisy examples. Experiments show that this joint spatio-temporal learnable filter consis

Load-bearing premise

The method needs a training set containing both clean and noisy versions of the signal; if only noisy signals are available, there is no supervision signal and the claimed 'partial prior information' advantage disappears.

Editorial extensions

If this is right

  • On eight real-world time-varying graph datasets, JFRFFNet achieves higher output SNR than GFRFFNet and ten graph baselines, ranking first on five datasets and second on three.
  • The learned transform-order pair and filter coefficients replace the grid search and analytic covariance computations of classical JFRFT Wiener filtering, adapting to each graph and dataset.
  • Because the transform layer is fully differentiable, the architecture can be trained end-to-end with any differentiable loss, not just denoising objectives.
  • Joint learning of spatial (graph) and temporal (time series) features is shown to yield consistent gains over spatial-only filtering, demonstrating the value of the JFRFT domain.
  • The method requires only partial prior information in the sense that it uses the graph topology and a training set of clean/noisy pairs, rather than explicit noise statistics.

Reading between the lines

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

  • The 'partial prior information' claim is narrower than it appears: the training set still needs clean reference signals, so the method exchanges analytic prior knowledge for labeled data. A true partial-information version would need a self-supervised or blind denoising loss, which the paper does not explore.
  • The three-layer stacking suggests that repeated application of the JFRFT filter in different fractional domains may act like a small learnable filter bank; testing with more layers or with shared versus per-layer orders would clarify what the network actually learns.
  • The O(N^3) eigendecomposition cost is the main computational bottleneck; replacing it with a fast approximate GFRFT could make the method practical for large graphs without changing the architecture.
  • The learned transform orders initialized at (0.5, 0.5) could be compared across datasets to see whether certain graph types prefer specific fractional domains, offering a diagnostic tool for graph signal structure.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 JFRFFNet, a graph signal denoising method that embeds the joint time-vertex fractional Fourier transform (JFRFT) into a neural network. The transform order pair (α, β) and an element-wise filter matrix H are treated as learnable parameters and updated by backpropagation, while the network itself is structured as JFRFT-domain filtering followed by inverse JFRFT. The method is evaluated on eight real-world time-varying graph datasets and compared with GFRFFNet and ten graph neural network baselines. The authors claim that JFRFFNet significantly improves output SNR and that the approach works with only partial prior information.

Significance. If the performance claims are validated, the proposed architecture would be a useful, interpretable addition to graph signal denoising, particularly because it jointly captures spatial and temporal features through a fractional transform. The main strengths are the explicit formulation of the trainable JFRFT layer, the gradient updates (Eqs. (7)–(8)), and the application to multiple real datasets. However, the evaluation protocol currently does not support the central claims: the reported results are obtained by post-hoc selection of the best graph-shift matrix per dataset, no error bars or significance tests are given, and the 'partial prior information' claim is contradicted by the use of clean/noisy training pairs. These issues are load-bearing for the paper's headline conclusions.

major comments (3)
  1. [Section IV-B / Table I] The reported ranking is obtained by selecting, for each method and dataset, the best result among five graph-shift matrices ('we present the denoising results for the best-performing matrix'). This is ex-post test-set selection over a family of models, not the performance of a fixed algorithm. The margins over GFRFFNet and other baselines are therefore maxima over five choices of graph-shift operator, which undermines the claim of 'best in five datasets, second in three'. Please report the full matrix sweep with a fixed selection rule (e.g., validation-based selection), averaged results over matrices, or all per-matrix results. Also specify why the METR adjacency results were discarded.
  2. [Section III-B / Abstract] The Abstract and Conclusion claim that JFRFFNet enables denoising with 'only partial prior information', but Section III-B states that 'The training and validation sets contain both clean and noisy signals'. The loss is MSE against clean signals, so the method requires clean training data. This is not the partial-prior scenario motivated in the Introduction, where complete signal/noise statistics are unavailable. The claim is therefore structurally unsupported. Please either reframe the contribution as supervised denoising, or provide an experiment in which only partial/statistical prior information is available (e.g., training on noisy targets or using known signal/noise statistics).
  3. [Section IV-A] The experimental protocol is under-specified: no noise type, noise construction, number of independent runs, standard deviations, or significance tests are reported. Table I gives single SNR values, so the claimed 'significant improvements' (Abstract) cannot be assessed. This is particularly important because many reported differences are 1–3 dB, which may be within run-to-run variability. Please add error bars, significance tests, and a detailed description of the noise-injection procedure.
minor comments (5)
  1. [General] There are typos and grammatical issues, e.g., 'acheives' in the Conclusion. The paper would benefit from careful proofreading.
  2. [Table I] Table I is very hard to read: multiple rows per dataset are merged into a single line, making it difficult to identify which SNR level corresponds to which baseline. Please reformat the table with clear row separation.
  3. [Equations (1) and (3)] The symbol D is used both for the temporal feature dimension (e.g., D in X ∈ R^{N×D}) and for the differentiation operator in the hyper-differential definitions (D^2). This is confusing; please use different notation for the operator.
  4. [Section III-B] The relationship between the vectorized gradient update in Eqs. (7)–(8) and the actual forward computation X_1 = F_G^α X (F^β)^⊤ should be spelled out explicitly, including the vectorization convention and how the filter matrix H is initialized and updated. This would improve reproducibility.
  5. [References] Reference [15] is the authors' own prior work and is central to the JFRFT layer. The letter should clearly state what is new relative to [15] and provide a more explicit comparison, since the current text reads as an application of that work.

Circularity Check

0 steps flagged · score 2.0 of 10

No derivation-level circularity; the reported SNR gains are empirical comparisons on held-out test data, with a minor self-citation of the trainable JFRFT layer and non-circular but notable overclaims about 'partial prior information' and best-matrix selection.

full rationale

Circularity analysis: The paper's central performance claim—JFRFFNet outperforming GFRFFNet and ten graph baselines in output SNR on eight datasets—is an empirical comparison on held-out test data against external baselines. It is not derived from the model equations, nor is any reported SNR computed as the value of a fitted parameter. The learned transform orders and filter coefficients are optimized on training pairs and then evaluated on test noisy signals, so the prediction is not the fit by construction. The main self-citation issue is in Section III-A: the trainable JFRFT layer and its gradient-update equations (7)–(8) are taken from the authors' own prior work [15] (Yan & Zhang, arXiv:2507.21527). This is a real self-citation and the layer is load-bearing for the network architecture, but it is not the source of the denoising comparison, so it does not make the performance claim circular. Two non-circular concerns are flagged explicitly. (i) The 'partial prior information' claim in the Abstract and Conclusion is unsupported: Section III-B states 'The training and validation sets contain both clean and noisy signals, while the test set contains only noisy signals,' which is full supervision, not partial prior information. This is a missing-support/overclaim, not a circular derivation. (ii) Section IV-B reports results 'for the best-performing matrix' and Fig. 2 shows all five matrices; choosing the best-performing matrix post hoc on the test set is a statistical validity threat that could inflate the reported ranking, but it is a selection-bias issue rather than an equation-level reduction of a prediction to an input. Overall, no step in the paper makes the output equal to the input by construction, so the circularity score is low; the minor self-citation and the two non-circular overclaims justify a score of 2 rather than 0.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The method's learnable parameters are the filter mask and the transform orders; these are fit to training data, so the reported test SNR is an evaluation of a fitted model. The clean-training-signal requirement is an unstated condition on the data. No new physical entities are introduced.

free parameters (3)
  • Filter coefficients H (N*D per layer) = learned, initialized to 1
    Element-wise mask in JFRFT domain, updated by backpropagation (Section III-B).
  • Transform order pairs (alpha, beta) per layer = initialized to (0.5, 0.5); learned
    JFRFT layer parameter updated via gradient descent, Eqs. (7)-(8).
  • Best graph matrix per dataset = chosen among adj, lap, rna, sna, nlap
    Table I reports the best-performing matrix for each dataset; used to maximize reported SNR (Section IV-B).
assumptions (3)
  • domain assumption The hyper-differential GFRFT and DFRFT definitions (Eqs. (1)-(3)) are valid, differentiable, and implementable as matrix exponentials.
    Borrowed from [12], [18]-[20]; the learnable layer and its gradients depend on this.
  • domain assumption The graph shift operator Z admits a Jordan decomposition and the JFRFT matrix is F_J^{alpha,beta} = F^beta (X) F_G^alpha.
    Defined in Section II-C following [13]-[15]; the whole architecture is built on this tensor product form.
  • domain assumption Clean training signals are available for every dataset.
    Section III-B states 'The training and validation sets contain both clean and noisy signals'; this is the supervision that makes the learned filter possible and contradicts the 'partial prior information' framing.

how reviews work

0 comments
Cite this review

Pith. "Pith review of JFRFFNet: A Data-Model Co-Driven Graph Signal Denoising Model with Partial Prior Information." pith.science (2026). https://pith.science/paper/JC53ZLDO

@misc{pith2026250909147,
  author       = {Pith},
  title        = {Pith review of: JFRFFNet: A Data-Model Co-Driven Graph Signal Denoising Model with Partial Prior Information},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JC53ZLDO}},
  note         = {Machine review of arXiv:2509.09147}
}
read the original abstract

Wiener filtering in the joint time-vertex fractional Fourier transform (JFRFT) domain has shown high effectiveness in denoising time-varying graph signals. Traditional filtering models use grid search to determine the transform-order pair and compute filter coefficients, while learnable ones employ gradient-descent strategies to optimize them; both require complete prior information of graph signals. To overcome this shortcoming, this letter proposes a data-model co-driven denoising approach, termed neural-network-aided joint time-vertex fractional Fourier filtering (JFRFFNet), which embeds the JFRFT-domain Wiener filter model into a neural network and updates the transform-order pair and filter coefficients through a data-driven approach. This design enables effective denoising using only partial prior information. Experiments demonstrate that JFRFFNet achieves significant improvements in output signal-to-noise ratio compared with some state-of-the-art methods.

Figures

Figures reproduced from arXiv: 2509.09147 by the authors.

Figure 1
Figure 1. Pipeline of the proposed JFRFFNet. In the feature filtering stage, the mapped features X1 are element￾wise multiplied by a trainable filter matrix H ∈ C N×D, with all the elements {hi,j} being learnable parameters, which adjust the weights of different components. The processing can be formulated as X2 = H ⊙ X1. Feature reconstruction is performed by the inverse JFRFT layer, which combines the inverse GFRFT and DFRF… view at source ↗
Figure 2
Figure 2. Denoising results for five different matrices, where [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 3 linked inside Pith

  1. [15]

    Trainable joint time-vertex fractional Fourier transform,

    Z. Yan and Z. Zhang, “Trainable joint time-vertex fractional Fourier transform,” 2025. [Online]. Available: https://arxiv.org/abs/2507.21527

    arXiv 2025

  2. [1]

    The emerging field of signal processing on graphs: Ex- tending high dimensional data analysis to networks and other irregular domains,

    D. I. Shuman, S. K. Narang, P. Frossard, A. Ortega, and P. Van- dergheynst, “The emerging field of signal processing on graphs: Ex- tending high dimensional data analysis to networks and other irregular domains,”IEEE Signal Process. Mag., vol. 30, pp. 83–98, May 2013

    2013

  3. [2]

    Big data analysis with signal processing on graphs: Representation and processing of massive data sets with irregular structure,

    A. Sandryhaila and J. M. F. Moura, “Big data analysis with signal processing on graphs: Representation and processing of massive data sets with irregular structure,”IEEE Signal Process. Mag., vol. 31, pp. 80–90, Sep. 2014

    2014

  4. [3]

    Graph signal denoising via trilateral filter on graph spectral domain,

    M. Onuki, S. Ono, M. Yamagishi, and Y . Tanaka, “Graph signal denoising via trilateral filter on graph spectral domain,”IEEE Trans. Signal Inf. Process. Netw., vol. 2, pp. 137–148, Jun. 2016

    2016

  5. [4]

    Graph signal processing – Part I: Graphs, graph spectra, and spectral clustering,

    L. Stankovic, D. Mandic, M. Dakovic, M. Brajovic, B. Scalzo, and T. Constantinides, “Graph signal processing – Part I: Graphs, graph spectra, and spectral clustering,” 2019. [Online]. Available: https://arxiv.org/abs/1907.03467

    arXiv 2019

  6. [5]

    Windowed fractional Fourier transform on graphs: Properties and fast algorithm,

    F. Yan and B. Li, “Windowed fractional Fourier transform on graphs: Properties and fast algorithm,”Digit. Signal Process., vol. 118, p. 103210, Nov. 2021

    2021

  7. [6]

    Windowed fractional Fourier transform on graphs: Fractional translation operator and Hausdorff-Young inequality,

    F. Yan, W. Gao, and B. Li, “Windowed fractional Fourier transform on graphs: Fractional translation operator and Hausdorff-Young inequality,” inProc. APSIPA Annu. Summit Conf. (APSIPA ASC), 2020, pp. 255–259

    2020

  8. [7]

    Spectral graph fractional Fourier transform for directed graphs and its application,

    F. Yan and B. Li, “Spectral graph fractional Fourier transform for directed graphs and its application,”Signal Process., vol. 210, p. 109099, Sep. 2023

    2023

Show all 40 references
  1. [8]

    Generalized sampling of graph signals with the prior information based on graph fractional Fourier transform,

    D. Wei and Z. Yan, “Generalized sampling of graph signals with the prior information based on graph fractional Fourier transform,”Signal Process., vol. 214, p. 109263, Jan. 2024

    2024

  2. [9]

    Hermitian random walk graph Fourier transform for directed graphs and its applications,

    D. Wei and S. Yuan, “Hermitian random walk graph Fourier transform for directed graphs and its applications,”Digit. Signal Process., vol. 155, p. 104751, Dec. 2024

    2024

  3. [10]

    The fractional Fourier transform on graphs,

    Y . Wang, B. Li, and Q. Cheng, “The fractional Fourier transform on graphs,” inProc. APSIPA Annu. Summit Conf. (APSIPA ASC), 2017, pp. 105–110

    2017

  4. [11]

    Optimal fractional Fourier filtering for graph signals,

    C. Ozturk, H. M. Ozaktas, S. Gezici, and A. Koc ¸, “Optimal fractional Fourier filtering for graph signals,”IEEE Trans. Signal Process., vol. 69, pp. 2902–2912, 2021

    2021

  5. [12]

    Graph fractional Fourier transform: A unified theory,

    T. Alikas ¸ifo˘glu, B. Kartal, and A. Koc ¸, “Graph fractional Fourier transform: A unified theory,”IEEE Trans. Signal Process., vol. 72, pp. 3834–3850, 2024

    2024

  6. [13]

    Joint time-vertex fractional Fourier transform,

    T. Alikas ¸ifo˘glu, B. Kartal, E. ¨Ozg¨unay, and A. Koc ¸, “Joint time-vertex fractional Fourier transform,”Signal Process., vol. 233, p. 109944, 2025

    2025

  7. [14]

    Wiener filtering in joint time- vertex fractional Fourier domains,

    T. Alikas ¸ifo˘glu, B. Kartal, and A. Koc ¸, “Wiener filtering in joint time- vertex fractional Fourier domains,”IEEE Signal Process. Lett., vol. 31, pp. 1319–1323, 2024

    2024

  8. [16]

    Trainable fractional Fourier transform,

    E. Koc ¸, T. Alikas ¸ifo˘glu, A. C. Aras, and A. Koc ¸, “Trainable fractional Fourier transform,”IEEE Signal Process. Lett., vol. 31, pp. 751–755, 2024

    2024

  9. [17]

    Discrete signal processing on graphs,

    A. Sandryhaila and J. Moura, “Discrete signal processing on graphs,” IEEE Trans. Signal Process., vol. 61, pp. 1644–1656, Apr. 2013

    2013

  10. [18]

    Operator theory-based computation of linear canonical transforms,

    A. Koc ¸ and H. M. Ozaktas, “Operator theory-based computation of linear canonical transforms,”Signal Process., vol. 189, p. 108291, Dec. 2021

    2021

  11. [19]

    Discrete linear canonical transform based on hyperdifferential operators,

    A. Koc ¸, B. Bartan, and H. M. Ozaktas, “Discrete linear canonical transform based on hyperdifferential operators,”IEEE Trans. Signal Process., vol. 67, pp. 2237–2248, May 2019

    2019

  12. [20]

    H. M. Ozaktas, Z. Zalevsky, and M. A. Kutay,The Fractional Fourier Transform with Applications in Optics and Signal Processing. New York, NY , USA: Wiley, 2001

    2001

  13. [21]

    Reconstruction of time-varying graph signals via Sobolev smoothness,

    J. Giraldo, A. Mahmood, B. Garcia-Garcia, D. Thanou, and T. Bouw- mans, “Reconstruction of time-varying graph signals via Sobolev smoothness,”IEEE Trans. Signal Inf. Process. Netw., vol. 8, pp. 201– 214, 2022

    2022

  14. [22]

    Stationary signal processing on graphs,

    N. Perraudin and P. Vandergheynst, “Stationary signal processing on graphs,”IEEE Trans. Signal Process., vol. 65, no. 13, pp. 3462–3477, Jul. 2017

    2017

  15. [23]

    SVD-based graph Fourier transforms on directed product graphs,

    C. Cheng, Y . Chen, Y . J. Lee, and Q. Sun, “SVD-based graph Fourier transforms on directed product graphs,”IEEE Trans. Signal Inf. Process. Netw., vol. 9, pp. 531–541, 2023

    2023

  16. [24]

    When spatio-temporal meet wavelets: Disentangled traffic forecasting via efficient spectral graph attention networks,

    Y . Fang, Y . Qin, H. Luo, F. Zhao, B. Xu, L. Zeng, and C. Wang, “When spatio-temporal meet wavelets: Disentangled traffic forecasting via efficient spectral graph attention networks,” inProc. IEEE Int. Conf. Data Eng. (ICDE), 2023, pp. 517–529

    2023

  17. [25]

    Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting,

    H. Wu, J. Xu, J. Wang, and M. Long, “Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting,” in Proc. NeurIPS, 2021, pp. 22 419–22 430

    2021

  18. [26]

    Diffusion convolutional recurrent neural network: Data-driven traffic forecasting,

    Y . Li, R. Yu, C. Shahabi, and Y . Liu, “Diffusion convolutional recurrent neural network: Data-driven traffic forecasting,” inProc. ICLR, 2018

    2018

  19. [27]

    Spectral temporal graph neural network for multivariate time-series forecasting,

    D. Cao, Y . Wang, J. Duan, C. Zhang, X. Zhu, C. Huang, Y . Tong, B. Xu, J. Bai, J. Tong, and Q. Zhang, “Spectral temporal graph neural network for multivariate time-series forecasting,” inProc. NeurIPS, vol. 33, 2020, pp. 17 766–17 778

    2020

  20. [28]

    Modeling long- and short- term temporal patterns with deep neural networks,

    G. Lai, W.-C. Chang, Y . Yang, and H. Liu, “Modeling long- and short- term temporal patterns with deep neural networks,” inProc. ACM SIGIR Conf. Res. Dev. Inf. Retr. (SIGIR). ACM, 2018, pp. 95–104

    2018

  21. [29]

    Frequency-domain MLPs are more effective learners in time series forecasting,

    K. Yi, Q. Zhang, W. Fan, S. Wang, P. Wang, H. He, N. An, D. Lian, L. Cao, and Z. Niu, “Frequency-domain MLPs are more effective learners in time series forecasting,” inProc. NeurIPS, vol. 36, 2023, pp. 76 656–76 679

    2023

  22. [30]

    Semi-supervised classification with graph convolutional networks,

    T. Kipf and M. Welling, “Semi-supervised classification with graph convolutional networks,” inProc. ICLR, 2017

    2017

  23. [31]

    Graph attention networks,

    P. Veli ˇckovi´c, G. Cucurull, A. Casanova, A. Romero, P. Li `o, and Y . Bengio, “Graph attention networks,” 2018. [Online]. Available: https://arxiv.org/abs/1710.10903

    2018 arXiv

  24. [32]

    Predict then propagate: Graph neural networks meet personalized pagerank,

    J. Gasteiger, A. Bojchevski, and S. G ¨unnemann, “Predict then propagate: Graph neural networks meet personalized pagerank,” inProc. ICLR, 2019

    2019

  25. [33]

    Graph neural networks with convolutional arma filters,

    F. M. Bianchi, D. Grattarola, L. Livi, and C. Alippi, “Graph neural networks with convolutional arma filters,”IEEE Trans. Pattern Anal. Mach. Intell., vol. 44, no. 7, pp. 3496–3507, Jul. 2022

    2022

  26. [34]

    Convolutional neural networks on graphs with fast localized spectral filtering,

    M. Defferrard, X. Bresson, and P. Vandergheynst, “Convolutional neural networks on graphs with fast localized spectral filtering,” inProc. NeurIPS, 2016, pp. 3844–3852

    2016

  27. [35]

    BernNet: Learning arbitrary graph spectral filters via bernstein approximation,

    M. He, Z. Wei, H. Xuet al., “BernNet: Learning arbitrary graph spectral filters via bernstein approximation,” inProc. NeurIPS, vol. 34, 2021, pp. 14 239–14 251

    2021

  28. [36]

    How powerful are spectral graph neural networks,

    X. Wang and M. Zhang, “How powerful are spectral graph neural networks,” inProc. ICML, vol. 162, 2022, pp. 23 341–23 362

    2022

  29. [37]

    Equivariant and stable positional encoding for more powerful graph neural networks,

    H. Wang, H. Yin, M. Zhang, and P. Li, “Equivariant and stable positional encoding for more powerful graph neural networks,” inProc. ICLR, 2022

    2022

  30. [38]

    LanczosNet: Multi-scale deep graph convolutional networks,

    R. Liao, Z. Zhao, R. Urtasun, and R. S. Zemel, “LanczosNet: Multi-scale deep graph convolutional networks,” inProc. ICLR, 2019

    2019

  31. [39]

    Specformer: Spectral graph neural networks meet transformers,

    D. Bo, C. Shi, L. Wang, and R. Liao, “Specformer: Spectral graph neural networks meet transformers,” inProc. ICLR, 2023

    2023

  32. [40]

    Masked label prediction: Unified message passing model for semi-supervised classification,

    Y . Shi, Z. Huang, S. Feng, H. Zhong, W. Wang, and Y . Sun, “Masked label prediction: Unified message passing model for semi-supervised classification,” inProc. ICLR, 2021

    2021

Pith tools

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