Pith. sign in

REVIEW 5 major objections 8 minor 3 cited by

S2FGL: Spatial Spectral Federated Graph Learning

T0 review · 5 major / 8 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read S2FGL claims that subgraph federated learning fails in two overlooked ways—lost label signals and mismatched frequency spectra—and reports that its two targeted losses fix both, beating all baselines on six datasets.

desk verdict Modest gains and clean motivation, but the missing spectral baselines (FedSSP and FedSPA) undercut the 'consistent outperformance' claim until added. read the letter →

arxiv 2507.02409 v4 pith:MGZFBBED submitted 2025-07-03 cs.LG cs.AI

Zihan Tan , Suyuan Huang , Guancheng Wan , Wenke Huang , He Li , Mang Ye This is my paper
classification cs.LGcs.AI
keywords FederatedgraphlearningSubgraphneuralnetworksSpectralheterogeneityLabelsignaldisruptionKnowledgedistillationspectrumNodeclassification
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

This paper claims that subgraph federated learning suffers from two distinct, previously unaddressed failures in how graph signals propagate: nodes lose label signals from former neighbors when edges are cut, and subgraphs have mismatched frequency spectra that push local models toward divergent signal-propagation schemes. To fix the first, it builds a global repository of class prototypes from structurally and semantically central nodes and distills that knowledge into local models. To fix the second, it reconstructs each subgraph's spectrum from node-feature similarities and aligns local high- and low-frequency projections with the global model. Across six homophilic and heterophilic datasets, the combined method reports accuracy gains of 0.9 to 2.0 points over FedAvg, with both components contributing in ablations.

What carries the argument

The argument runs on two loss terms appended to the standard cross-entropy objective. NLIR uses a Structure-Aware Label Centrality (SALC) metric, which combines Personalized PageRank structural prominence with label-influence diffusion, to select representative nodes; local prototypes are aggregated into four global anchor prototypes per class, and a KL-divergence loss aligns the cosine-similarity distribution of local and frozen global features over those prototypes. FGMA constructs a sparse $k$-nearest-neighbour cosine-similarity matrix $S'$ from node features, forms the Laplacian $L' = D' - S'$, eigendecomposes it, projects local and global features onto the chosen lowest- and highest-frequency eigenvectors, and minimizes MSE between the paired projections. Together they constitute the S2FGL objective $L_{CE} + \lambda_1 L_{FKD} + \lambda_2 L_{FGMA}$, where the first loss injects semantic knowledge and the second aligns spectral behavior.

What would settle it

Randomly permute node features within each client before training while keeping the true graph fixed: if S2FGL still beats FedAvg, FGMA's improvement is not coming from spectral alignment on a faithful reconstruction. A direct check is to measure edge overlap or spectral distance between the learned similarity matrix and the true adjacency matrix and see whether FGMA's gain tracks that fidelity.

Watch

Extended reading notes

Core claim

The paper's central claim is that existing subgraph-FL methods, which focus on structural collaboration, miss the graph-signal dynamics that actually degrade the global model. It identifies label signal disruption: when edges are cut, nodes lose label signals from formerly adjacent labeled nodes, lowering the Structure Inertia Score and weakening semantic knowledge; and spectral client drift: heterogeneity in eigenvalue distributions across clients makes local GNNs overfit their own frequency responses. S2FGL addresses both with two losses—NLIR selects top-K nodes by a structural-and-label centrality, builds a global prototype repository, and uses KL-divergence to align local and global similarity distributions over those prototypes; FGMA builds a sparse cosine-similarity graph from features, takes its Laplacian's low- and high-frequency eigenvectors, projects features onto them, and aligns local and global projections with MSE. The paper reports that S2FGL consistently outperforms all baselines on all six datasets, with gains of 1.5, 1.7, and 1.3 points on Cora, Citeseer, and PubMed, and 2.0, 1.4, and 0.9 points on Texas, Wisconsin, and Minesweeper.

Load-bearing premise

The load-bearing assumption is that the feature-based $k$-nearest-neighbour cosine similarity matrix faithfully stands in for the real graph's adjacency, so the eigenvectors of its Laplacian describe how the GNN actually propagates signals; if node features are noisy or uninformative, the frequency alignment may be optimizing against a graph that does not match reality.

Editorial extensions

If this is right

  • S2FGL reports accuracy gains over FedAvg on every dataset tested: +1.5 on Cora, +1.7 on Citeseer, +1.3 on PubMed, +2.0 on Texas, +1.4 on Wisconsin, and +0.9 on Minesweeper.
  • Ablations on Cora and Citeseer show both losses matter: NLIR alone gives 83.2 and 75.6, FGMA alone gives 82.6 and 75.0, and the full model gives 83.4 and 76.0, versus FedAvg's 81.9 and 74.3.
  • The targeted-effectiveness analysis predicts that the methods help most exactly where the diagnosed problems are worst: NLIR's gain grows as the structure inertia score falls, and FGMA's gain grows with inter-client spectral heterogeneity.
  • Reported performance stays stable across varying hyperparameters (prototype counts and alignment weights) and client scales of 5, 10, and 20.
  • Because the two losses target different failure modes, combining them with structural neighbor-generation methods is a natural extension for even more fragmented subgraph scenarios.

Reading between the lines

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

  • Beyond the paper, FGMA's mechanism predicts a falsifiable boundary: on graphs whose raw features are uninformative or deliberately randomized, the cosine-similarity spectrum breaks down, so the reported gain should shrink or disappear.
  • Beyond the paper, the same frequency-alignment recipe could transfer to non-graph federated learning, where any learned representation defines an implicit similarity graph; aligning low- and high-frequency projections may reduce client drift there, though this is untested.
  • Beyond the paper, the global prototype repository offers a privacy-facing design choice: sharing only class-level anchor prototypes avoids transmitting edges or raw features, but the paper does not quantify the privacy-utility trade-off against direct gradient exchange.
  • If the 0.9-to-2.0 point gains hold at larger scale, NLIR and FGMA could serve as drop-in regularizers for subgraph-FL systems, since they make no change to the federated aggregation protocol itself.
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

5 major / 8 minor

Summary. This paper studies subgraph federated graph learning (subgraph-FL) and identifies two phenomena: label signal disruption caused by edge loss across clients, and spectral client drift caused by spectral heterogeneity across subgraphs. To address these, it proposes S2FGL with two auxiliary losses: NLIR, which builds a global prototype repository from structurally and semantically representative nodes and aligns local/global similarity distributions via KL divergence, and FGMA, which reconstructs a Laplacian from feature similarity, projects features onto low/high-frequency eigenspaces, and aligns local with global projections via MSE. Experiments on six datasets (Cora, Citeseer, Pubmed, Texas, Wisconsin, Minesweeper) with the ACM-GCN backbone report accuracy gains over FedAvg of 0.9 to 2.0 points and claim consistent superiority over ten baselines. The code is publicly released.

Significance. If the claims hold, the paper makes a modest but useful contribution to subgraph-FL: it provides an explicit spatial and spectral motivation and two losses that improve global accuracy over FedAvg and several FGL baselines, with ablations showing each component contributes (Table 2) and analyses relating gains to SIS and spectral KL divergence (Figure 6). The empirical motivation, public code release, and evaluation across homophilic and heterophilic datasets are strengths. However, the strength of the central claim depends on the completeness of the baseline set and on reproducibility details: the two most directly relevant spectral subgraph-FL baselines are absent, and several hyperparameters are unspecified. With those gaps closed, the contribution would be a solid, if incremental, addition to the subgraph-FL literature.

major comments (5)
  1. [Section 5.1, Table 1] Section 5.1 lists ten baselines, but Table 1 omits FedSSP (Tan et al., 2024) and FedSPA (Tan et al., 2025), the two most directly related spectral/topology-aware subgraph-FL methods cited in Section 2. Because Section 5.2 Q1 asserts that S2FGL 'consistently outperforms all baseline approaches across all six datasets,' the absence of these comparators leaves the central claim unverified; please add them under the same protocol or explicitly restrict the claim to the evaluated baselines.
  2. [Section 5.2 Q3, Sections 4.2-4.3] The values of lambda_1, lambda_2, k_sim, and k_eig used for Table 1 are never stated: Section 4.2 defines K but not the number of anchor prototypes per class, and Section 5.2 Q3 reports only that NLIR hyperparameters were tested over '100, 50, 10, and 1' and FGMA over '0.01, 0.05, 0.5, and 1' without naming which parameter took which values. Without these defaults, the main results are not reproducible; please specify the full configuration for every experiment.
  3. [Section 2] Section 2 states 'we are the first to empirically reveal and address the challenge of poor semantic knowledge under label signal disruption and client drift under spectral heterogeneity among subgraphs,' yet FedSSP, cited in the same section, explicitly addresses spectral knowledge and client preference in subgraph-FL. This overstates novelty and is inconsistent with the paper's own related work; please either provide a concrete technical differentiation from FedSSP and FedSPA or soften the claim.
  4. [Section 4.3, Eqs. (9)-(12)] FGMA reconstructs the graph Laplacian from a k_sim-nearest-neighbor cosine similarity matrix S' of node features and aligns projections on this reconstructed spectrum, but the paper gives no evidence that S' approximates the original adjacency structure or that the eigen-projections reflect the GNN's true spectral behavior. If node features are noisy or non-discriminative, the alignment could be against a graph that carries no information about signal propagation; please add a validation, such as comparing spectra of L' and L or ablating with the true adjacency where permissible.
  5. [Table 1, Section 5.1] All experiments use a single backbone (ACM-GCN), and several reported gains over the strongest baseline are within one standard deviation (e.g., Cora 83.4±0.2 vs. FedFa 82.7±0.5; Minesweeper 80.5±0.1 vs. FGGP 80.4±0.3), with no significance tests reported. Since the central claim is consistent superiority, please report paired significance tests or confidence intervals over the five runs and, if feasible, verify the main comparison with at least one additional backbone.
minor comments (8)
  1. [Throughout] Please correct typos: 'Conequently' in Section 4.3, 'NILR' in Section 4.2 and Figure 6, and 'experimantal' in Section 5.1.
  2. [Section 4.2, Eq. (7)] The notation alpha is used both for the PPR teleportation probability in Eqs. (1) and (5) and for 'alpha percent of the clients' in the description of Eq. (7); use distinct symbols for these two quantities.
  3. [Section 4.2, Eq. (7)] Eq. (7) says four global anchor prototypes are constructed for each class, but the text does not explain how this number was chosen or whether it is swept; please clarify.
  4. [Section 5.2 Q3, Figures 4-5] Figures 4 and 5 should identify which hyperparameter is varied and which values are held fixed; the current captions do not allow the reader to interpret the plots.
  5. [Section 5.1] Section 5.1 should state the number of communication rounds and local epochs used for Table 1; reporting only the average over the last five rounds is insufficient for reproduction.
  6. [Definition 3.2] SIS as defined sums unnormalized maximum PPR values; please state whether a normalized variant is used so that comparisons across different client scales are meaningful.
  7. [References] References contain venue-abbreviation and spelling errors, including 'ASTAT17' (should be 'AISTATS17') and 'NeruIPS' in several entries; please unify the abbreviations.
  8. [Related Work] The heading 'Graph Spectrum' in Section 2 is a grammatical fragment; consider rewriting it as 'Graph Spectrum Learning' or a complete sentence.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: S2FGL's central accuracy claim is independently measured on held-out test sets, and its proposed losses do not optimize the diagnostic metrics used for motivation.

full rationale

The paper contains no first-principles derivation whose output reduces to its input by construction. The two diagnostic quantities, SIS (Eq. 2) and the KL divergence of eigenvalue distributions used in Fig. 1, are defined by the authors and used to motivate NLIR and FGMA. However, neither proposed loss (Eq. 8 for LFKD and Eq. 12 for LFGMA) directly optimizes SIS or the eigenvalue-distribution KL, and the central empirical claim in Q1 is measured as node-classification accuracy on held-out test splits. The ablation and correlation studies (Fig. 6) also evaluate performance gains in accuracy, not improvement in the motivating metrics themselves. The paper does cite prior work by the same authorship group, including FedSSP and FedSPA, and the claim that the spectral domain in subgraph-FL is 'unexplored' is questionable given those citations. That is an evaluation-completeness and novelty concern, not a circularity concern: no conclusion in the method section is derived from a self-citation, and the omitted baselines do not make the reported test accuracy equivalent to any fitted input. The honest finding is therefore no significant circularity.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The paper does not introduce new physical entities; the global knowledge repository is a server-side data structure. The free parameters listed are those whose values affect the reported results but are not fully specified in the text, especially lambda_1, lambda_2, k_sim, and k_eig. The domain assumptions are proxies for the two named phenomena, and the first two proxies are defined in the same paper, which slightly raises the circularity burden of the motivation section.

free parameters (7)
  • K = 1/3 of total nodes
    Number of representative nodes selected per client for prototype computation. Set by convention, not justified in text.
  • k_sim
    Number of nearest feature neighbors used to construct the similarity matrix in Eq. (9); no default value is given.
  • k_eig
    Number of low and high frequency eigenvectors used in FGMA; no default value is given.
  • lambda_1
    Weight of the LFKD loss; sensitivity is tested over 1-100 in Fig. 4 but the value used in Table 1 is not reported.
  • lambda_2
    Weight of the LFGMA loss; sensitivity is tested over 0.01-1 in Fig. 4 but the value used in Table 1 is not reported.
  • n_proto = 4
    Number of global anchor prototypes per class in Eq. (7); the choice of four is not justified.
  • alpha (PPR teleportation) = 0.15
    Standard PPR teleportation probability, a hand-set constant from prior literature. The same symbol is reused in Sec. 4.2 for the percentage of clients aggregated into each global prototype, where its value is ambiguous.
assumptions (4)
  • domain assumption Louvain community detection produces a realistic and valid subgraph decomposition for subgraph-FL.
    Used in Section 5.1 to create client subgraphs; the representativeness of this partitioning for real federated graph settings is assumed without discussion.
  • ad hoc to paper SIS (Eq. 2) faithfully quantifies label influence on graph structure, so its decrease indicates a loss of semantic knowledge.
    Introduced in Definition 3.2 and used as evidence in Fig. 1; no external validation that SIS correlates with downstream GNN accuracy.
  • ad hoc to paper The KL divergence between per-client eigenvalue distributions is a valid measure of spectral heterogeneity that degrades collaboration.
    Used in Fig. 1 and Fig. 6 to motivate FGMA; the causal link from this divergence to accuracy loss is not established.
  • domain assumption The top-k_sim cosine similarity graph approximates the GNN's adjacency awareness, and its spectrum is a meaningful alignment target.
    Eqs. (9)-(11) reconstruct spectra from node features; this presumes that feature similarity captures the structure relevant to signal propagation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of S2FGL: Spatial Spectral Federated Graph Learning." pith.science (2026). https://pith.science/paper/MGZFBBED

@misc{pith2026250702409,
  author       = {Pith},
  title        = {Pith review of: S2FGL: Spatial Spectral Federated Graph Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MGZFBBED}},
  note         = {Machine review of arXiv:2507.02409}
}
read the original abstract

Federated Graph Learning (FGL) combines the privacy-preserving capabilities of federated learning (FL) with the strong graph modeling capability of Graph Neural Networks (GNNs). Current research addresses subgraph-FL from the structural perspective, neglecting the propagation of graph signals on spatial and spectral domains of the structure. From a spatial perspective, subgraph-FL introduces edge disconnections between clients, leading to disruptions in label signals and a degradation in the semantic knowledge of the global GNN. From a spectral perspective, spectral heterogeneity causes inconsistencies in signal frequencies across subgraphs, which makes local GNNs overfit the local signal propagation schemes. As a result, spectral client drift occurs, undermining global generalizability. To tackle the challenges, we propose a global knowledge repository to mitigate the challenge of poor semantic knowledge caused by label signal disruption. Furthermore, we design a frequency alignment to address spectral client drift. The combination of Spatial and Spectral strategies forms our framework S2FGL. Extensive experiments on multiple datasets demonstrate the superiority of S2FGL. The code is available at https://github.com/Wonder7racer/S2FGL.git.

Figures

Figures reproduced from arXiv: 2507.02409 by the authors.

Figure 1
Figure 1. In the first place, compared with centralized GNN train￾ing, subgraph-FL is encountering label signal disruption chal￾lenge, leading to decreased structure inertia score and poor se￾mantic knowledge for GNNs. Moreover, we demonstrate the heat map of the Kullback-Leibler divergence of eigenvalue distributions across clients. Inconsistency in subgraph signal frequency caused by spectral heterogeneity leads to spectral… view at source ↗
Figure 2
Figure 2. Problem Illustration. (a) From the spatial perspective, nodes in subgraph-FL lose label signals from originally nearby labeled nodes due to edge loss, namely label signal disruption. Correspondingly, GNNs suffer from poor semantic knowledge, leading to a deteriorated global GNN. (b) From the spectral perspective, spectral heterogeneity induces inconsistencies in signal frequencies across subgraphs, leading to spectr… view at source ↗
Figure 3
Figure 3. Framework Illustration. (a) Node Label Information Reinforcement (NLIR) leverages a structurally and semantically representative global prototype repository. It provides multidimensional semantic localization of nodes through similarity distribution and allows LFKD to inject the semantic knowledge during training. (b) Frequency-aware Graph Modeling Alignment (FGMA) aligns local high and low spectral adjacency awaren… view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: Analysis of performance under different client scales. and adaptability across varying hyperparameter settings and client partition configurations on both the Cora and Cite￾seer datasets. These findings confirm that S 2FGL not only sustains its effectiveness under dive…
Figure 4
Figure 4. Figure 4: Analysis of the performance growth between S 2 FGL and FedAvg under different hyperparameters of NLIR and FGMA Varying Client Scales. We assessed performance with dif￾ferent client scales: 5, 10 and 20. Specifically, we compared S 2FGL with other FL and FGL baselines, …

Discussion (0). Sign in to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. STAGE: Tackling Semantic Drift in Multimodal Federated Graph Learning

    cs.LG 2026-05 unverdicted novelty 7.0 of 10

    STAGE builds a shared semantic space through feature translation and controlled graph propagation to reduce semantic drift in multimodal federated graph learning, delivering state-of-the-art results with lower communi...

  2. Towards Effective Federated Multimodal Graph Learning via Navigating Multifaceted Heterogeneity

    cs.LG 2026-08 conditional novelty 6.0 of 10

    A federated multimodal graph-learning method that uses server-side routing of topology-aware prototypes to align clients across tasks, modalities, and topologies, outperforming baselines on 8 datasets.

  3. PRISM: Topology-Aware Cross-Modal Imputation for Modality-Deficient Federated Graph Learning

    cs.LG 2026-06 unverdicted novelty 5.0 of 10

    PRISM proposes a topology-aware cross-modal imputation framework for client-level modality-deficient multimodal federated graph learning that improves deficient clients by 4.48% on average over baselines across six datasets.

Reference graph

Works this paper leans on

77 extracted references · 70 canonical work pages · cited by 3 Pith papers

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    Acar, D. A. E., Zhao, Y., Matas, R., Mattina, M., Whatmough, P., and Saligrama, V. Federated learning based on dynamic regularization. In ICLR, 2021

    work page 2021

  3. [3]

    Baek, J., Jeong, W., Jin, J., Yoon, J., and Hwang, S. J. Personalized subgraph federated learning. In ICML, pp.\ 1396--1415, 2023

    work page 2023

  4. [4]

    A critical look at the evaluation of gnns under heterophily: Are we really making progress? arXiv preprint arXiv:2302.11640, 2023

    Baranovskiy, D., Oseledets, I., and Babenko, A. A critical look at the evaluation of gnns under heterophily: Are we really making progress? arXiv preprint arXiv:2302.11640, 2023

    arXiv 2023

  5. [5]

    Graph contrastive learning with stable and scalable spectral encoding

    Bo, D., Fang, Y., Liu, Y., and Shi, C. Graph contrastive learning with stable and scalable spectral encoding. NeurIPS, 36: 0 45516--45532, 2023 a

    work page 2023

  6. [6]

    Specformer: Spectral graph neural networks meet transformers

    Bo, D., Shi, C., Wang, L., and Liao, R. Specformer: Spectral graph neural networks meet transformers. arXiv preprint arXiv:2303.01028, 2023 b

    arXiv 2023

  7. [7]

    and Chao, W.-L

    Chen, H.-Y. and Chao, W.-L. On bridging generic and personalized federated learning for image classification. In ICLR, 2022

    work page 2022

  8. [8]

    Fede: Embedding knowledge graphs in federated setting

    Chen, M., Zhang, W., Yuan, Z., Jia, Y., and Chen, H. Fede: Embedding knowledge graphs in federated setting. In IJCKG, pp.\ 80--88, 2021

    work page 2021

Show all 77 references
  1. [9]

    Fair federated learning under domain skew with local consistency and domain diversity

    Chen, Y., Huang, W., and Ye, M. Fair federated learning under domain skew with local consistency and domain diversity. In CVPR, pp.\ 12077--12086, 2024

    2024

  2. [10]

    Learning to extract symbolic knowledge from the world wide web

    Craven, M., DiPasquo, D., Freitag, D., McCallum, A., Mitchell, T., Nigam, K., and Slattery, S. Learning to extract symbolic knowledge from the world wide web. AAAI/IAAI, pp.\ 2, 1998

    1998

  3. [11]

    Convolutional neural networks on graphs with fast localized spectral filtering

    Defferrard, M., Bresson, X., and Vandergheynst, P. Convolutional neural networks on graphs with fast localized spectral filtering. In NeurIPS, 2016

    2016

  4. [12]

    H., Yan, S., He, C., Ferrara, E., and Avestimehr, A

    Ezzeldin, Y. H., Yan, S., He, C., Ferrara, E., and Avestimehr, A. S. Fairfed: Enabling group fairness in federated learning. In AAAI, 2023

    2023

  5. [13]

    A graph neural network framework for social recommendations

    Fan, W., Ma, Y., Li, Q., Wang, J., Cai, G., Tang, J., and Yin, D. A graph neural network framework for social recommendations. TKDE, 2020

    2020

  6. [14]

    and Ye, M

    Fang, X. and Ye, M. Robust federated learning with noisy and heterogeneous clients. In CVPR, pp.\ 10072--10081, 2022

    2022

  7. [15]

    Fang, X., Easwaran, A., Genest, B., and Suganthan, P. N. Adaptive hierarchical graph cut for multi-granularity out-of-distribution detection. IEEE TAI, 2025

    2025

  8. [16]

    Federated graph machine learning: A survey of concepts, techniques, and applications

    Fu, X., Zhang, B., Dong, Y., Chen, C., and Li, J. Federated graph machine learning: A survey of concepts, techniques, and applications. arXiv preprint arXiv:2207.11812, 2022

    2022 arXiv

  9. [17]

    Addressing heterophily in graph anomaly detection: A perspective of graph spectrum

    Gao, Y., Wang, X., He, X., Liu, Z., Feng, H., and Zhang, Y. Addressing heterophily in graph anomaly detection: A perspective of graph spectrum. In Proceedings of the ACM Web Conference 2023, pp.\ 1528--1538, 2023

    2023

  10. [18]

    L., Bollacker, K

    Giles, C. L., Bollacker, K. D., and Lawrence, S. Citeseer: An automatic citation indexing system. In Proceedings of the third ACM conference on Digital libraries, pp.\ 89--98, 1998

    1998

  11. [19]

    Structural fairness-aware active learning for graph neural networks

    Han, H., Liu, X., Ma, L., Torkamani, M., Liu, H., Tang, J., and Yamada, M. Structural fairness-aware active learning for graph neural networks. In ICLR, 2023

    2023

  12. [20]

    S., Rong, Y., et al

    He, C., Balasubramanian, K., Ceyani, E., Yang, C., Xie, H., Sun, L., He, L., Yang, L., Yu, P. S., Rong, Y., et al. Fedgraphnn: A federated learning system and benchmark for graph neural networks. In ICLR, 2021 a

    2021

  13. [21]

    Bernnet: Learning arbitrary graph spectral filters via bernstein approximation

    He, M., Wei, Z., Xu, H., et al. Bernnet: Learning arbitrary graph spectral filters via bernstein approximation. In NeurIPS, pp.\ 14239--14251, 2021 b

    2021

  14. [22]

    Convolutional neural networks on graphs with chebyshev approximation, revisited

    He, M., Wei, Z., and Wen, J.-R. Convolutional neural networks on graphs with chebyshev approximation, revisited. In NeurIPS, pp.\ 7264--7276, 2022

    2022

  15. [23]

    Federated robustness propagation: sharing adversarial robustness in heterogeneous federated learning

    Hong, J., Wang, H., Wang, Z., and Zhou, J. Federated robustness propagation: sharing adversarial robustness in heterogeneous federated learning. In AAAI, pp.\ 7893--7901, 2023

    2023

  16. [24]

    Is aggregation the only choice? federated learning via layer-wise model recombination

    Hu, M., Yue, Z., Xie, X., Chen, C., Huang, Y., Wei, X., Lian, X., Liu, Y., and Chen, M. Is aggregation the only choice? federated learning via layer-wise model recombination. In SIGKDD, pp.\ 1096--1107, 2024

    2024

  17. [25]

    Learn from others and be yourself in heterogeneous federated learning

    Huang, W., Ye, M., and Du, B. Learn from others and be yourself in heterogeneous federated learning. In CVPR, pp.\ 10143--10153, 2022

    2022

  18. [26]

    Federated graph semantic and structural learning

    Huang, W., Wan, G., Ye, M., and Du, B. Federated graph semantic and structural learning. In Proceedings of the Thirty-Second International Joint Conference on Artificial Intelligence, pp.\ 3830--3838, 2023 a

    2023

  19. [27]

    Generalizable heterogeneous federated cross-correlation and instance similarity learning

    Huang, W., Ye, M., Shi, Z., and Du, B. Generalizable heterogeneous federated cross-correlation and instance similarity learning. TPAMI, pp.\ 712--728, 2023 b

    2023

  20. [28]

    Rethinking federated learning with domain shift: A prototype view

    Huang, W., Ye, M., Shi, Z., Li, H., and Du, B. Rethinking federated learning with domain shift: A prototype view. In CVPR, pp.\ 16312--16322. IEEE, 2023 c

    2023

  21. [29]

    A federated learning for generalization, robustness, fairness: A survey and benchmark

    Huang, W., Ye, M., Shi, Z., Wan, G., Li, H., Du, B., and Yang, Q. A federated learning for generalization, robustness, fairness: A survey and benchmark. TPAMI, 2024

    2024

  22. [30]

    P., Kale, S., Mohri, M., Reddi, S

    Karimireddy, S. P., Kale, S., Mohri, M., Reddi, S. J., Stich, S. U., and Suresh, A. T. Scaffold: Stochastic controlled averaging for on-device federated learning. In ICML, pp.\ 5132--5143, 2020

    2020

  23. [31]

    Rethinking graph transformers with spectral attention

    Kreuzer, D., Beaini, D., Hamilton, W., L \'e tourneau, V., and Tossou, P. Rethinking graph transformers with spectral attention. In NeruIPS, pp.\ 21618--21629, 2021

    2021

  24. [32]

    Preservation of the global knowledge by not-true distillation in federated learning

    Lee, G., Jeong, M., Shin, Y., Bae, S., and Yun, S.-Y. Preservation of the global knowledge by not-true distillation in federated learning. In Koyejo, S., Mohamed, S., Agarwal, A., Belgrave, D., Cho, K., and Oh, A. (eds.), NeurIPS, pp.\ 38461--38474, 2022

    2022

  25. [33]

    K., Zaheer, M., Sanjabi, M., Talwalkar, A., and Smith, V

    Li, T., Sahu, A. K., Zaheer, M., Sanjabi, M., Talwalkar, A., and Smith, V. Federated optimization in heterogeneous networks. MLSys, 2: 0 429--450, 2020

    2020

  26. [34]

    Fedgta: Topology-aware averaging for federated graph learning

    Li, X., Wu, Z., Zhang, W., Zhu, Y., Li, R.-H., and Wang, G. Fedgta: Topology-aware averaging for federated graph learning. arXiv preprint arXiv:2401.11755, 2024

    2024 arXiv

  27. [35]

    Fedphp: Federated personalization with inherited private models

    Li, X.-C., Zhan, D.-C., Shao, Y., Li, B., and Song, S. Fedphp: Federated personalization with inherited private models. In ECML, pp.\ 587--602, 2021

    2021

  28. [36]

    Liao, R., Zhao, Z., Urtasun, R., and Zemel, R. S. Lanczosnet: Multi-scale deep graph convolutional networks. In ICLR, 2019

    2019

  29. [37]

    Revisiting graph contrastive learning from the perspective of graph spectrum

    Liu, N., Wang, X., Bo, D., Shi, C., and Pei, J. Revisiting graph contrastive learning from the perspective of graph spectrum. In NeurIPS, pp.\ 2972--2983, 2022

    2022

  30. [38]

    and Yu, H

    Liu, R. and Yu, H. Federated graph neural networks: Overview, techniques and challenges. arXiv preprint arXiv:2202.07256, 2022

    2022 arXiv

  31. [39]

    Graph condensation via eigenbasis matching

    Liu, Y., Bo, D., and Shi, C. Graph condensation via eigenbasis matching. arXiv preprint arXiv:2310.09202, 2023

    2023 arXiv

  32. [40]

    A., Lau, M

    Liu, Z., Wan, G., Prakash, B. A., Lau, M. S., and Jin, W. A review of graph neural networks in epidemic modeling. arXiv preprint arXiv:2403.19852, 2024

    2024 arXiv

  33. [41]

    Revisiting heterophily for graph neural networks

    Luan, S., Hua, C., Lu, Q., Zhu, J., Zhao, M., Zhang, S., Chang, X.-W., and Precup, D. Revisiting heterophily for graph neural networks. In NeurIPS22, pp.\ 1362--1375, 2022

    2022

  34. [42]

    Personalized federated learning on heterogeneous and long-tailed data via expert collaborative learning

    Lv, F., Shang, X., Zhou, Y., Zhang, Y., Li, M., and Lu, Y. Personalized federated learning on heterogeneous and long-tailed data via expert collaborative learning. arXiv preprint arXiv:2408.02019, 2024

    2024 arXiv

  35. [43]

    K., Nigam, K., Rennie, J., and Seymore, K

    McCallum, A. K., Nigam, K., Rennie, J., and Seymore, K. Automating the construction of internet portals with machine learning. Information Retrieval, 3 0 (2): 0 127--163, 2000

    2000

  36. [44]

    McMahan, B., Moore, E., Ramage, D., Hampson, S., and y Arcas, B. A. Communication-efficient learning of deep networks from decentralized data. In AISTATS, pp.\ 1273--1282, 2017

    2017

  37. [45]

    Fairness in federated learning via core-stability

    Ray Chaudhury, B., Li, L., Kang, M., Li, B., and Mehta, R. Fairness in federated learning via core-stability. In NeurIPS, pp.\ 5738--5750, 2022

    2022

  38. [46]

    Collective classification in network data

    Sen, P., Namata, G., Bilgic, M., Getoor, L., Galligher, B., and Eliassi-Rad, T. Collective classification in network data. AI magazine, 29 0 (3): 0 93--93, 2008

    2008

  39. [47]

    Federated learning on heterogeneous and long-tailed data via classifier re-training with federated features

    Shang, X., Lu, Y., Huang, G., and Wang, H. Federated learning on heterogeneous and long-tailed data via classifier re-training with federated features. In IJCAI, 2022

    2022

  40. [48]

    Smith, V., Chiang, C.-K., Sanjabi, M., and Talwalkar, A. S. Federated multi-task learning. In NeurIPS, 2017

    2017

  41. [49]

    Federated learning on non-iid graphs via structural knowledge sharing

    Tan, Y., Liu, Y., Long, G., Jiang, J., Lu, Q., and Zhang, C. Federated learning on non-iid graphs via structural knowledge sharing. In AAAI, pp.\ 9953--9961, 2023

    2023

  42. [50]

    Fedssp: Federated graph learning with spectral knowledge and personalized preference

    Tan, Z., Wan, G., Huang, W., and Ye, M. Fedssp: Federated graph learning with spectral knowledge and personalized preference. In NeurIPS, 2024

    2024

  43. [51]

    Fedspa: Generalizable federated graph learning under homophily heterogeneity

    Tan, Z., Wan, G., Huang, W., Li, H., Zhang, G., Yang, C., and Ye, M. Fedspa: Generalizable federated graph learning under homophily heterogeneity. In CVPR, 2025

    2025

  44. [52]

    Rethinking graph neural networks for anomaly detection

    Tang, J., Li, J., Gao, Z., and Li, J. Rethinking graph neural networks for anomaly detection. In ICML, pp.\ 21076--21089, 2022

    2022

  45. [53]

    Federated graph learning under domain shift with generalizable prototypes

    Wan, G., Huang, W., and Ye, M. Federated graph learning under domain shift with generalizable prototypes. In AAAI, pp.\ 15429--15437, 2024 a

    2024

  46. [54]

    V., and Ye, M

    Wan, G., Tian, Y., Huang, W., Chawla, N. V., and Ye, M. S3gcl: Spectral, swift, spatial graph contrastive learning. In ICML, pp.\ 49973--49990, 2024 b

    2024

  47. [55]

    Rethink graphode generalization within coupled dynamical system

    Wan, G., Huang, Z., Zhao, W., Luo, X., Sun, Y., and Wang, W. Rethink graphode generalization within coupled dynamical system. In ICML, 2025 a

    2025

  48. [56]

    Energy-based backdoor defense against federated graph learning

    Wan, G., Shi, Z., Huang, W., Zhang, G., Tao, D., and Ye, M. Energy-based backdoor defense against federated graph learning. In ICLR, 2025 b

    2025

  49. [57]

    A semi-supervised graph attentive network for financial fraud detection

    Wang, D., Lin, J., Cui, P., Jia, Q., Wang, Z., Fang, Y., Yu, Q., Zhou, J., Yang, S., and Qi, Y. A semi-supervised graph attentive network for financial fraud detection. In ICDM, pp.\ 598--607, 2019

    2019

  50. [58]

    Wang, J., Liu, Q., Liang, H., Joshi, G., and Poor, H. V. Tackling the objective inconsistency problem in heterogeneous federated optimization. In NeurIPS, pp.\ 7611--7623, 2020

    2020

  51. [59]

    and Zhang, M

    Wang, X. and Zhang, M. How powerful are spectral graph neural networks. In ICML, pp.\ 23341--23362, 2023

    2023

  52. [60]

    Bold but cautious: Unlocking the potential of personalized federated learning through cautiously aggressive collaboration

    Wu, X., Liu, X., Niu, J., Zhu, G., and Tang, S. Bold but cautious: Unlocking the potential of personalized federated learning through cautiously aggressive collaboration. In ICCV, pp.\ 19375--19384, 2023

    2023

  53. [61]

    Wu, Z., Pan, S., Chen, F., Long, G., Zhang, C., and Philip, S. Y. A comprehensive survey on graph neural networks. TNNLS, pp.\ 4--24, 2020

    2020

  54. [62]

    Federated graph classification over non-iid graphs

    Xie, H., Ma, J., Xiong, L., and Yang, C. Federated graph classification over non-iid graphs. In NeruIPS, pp.\ 18839--18852, 2021

    2021

  55. [63]

    Asynchronous federated learning on heterogeneous devices: A survey

    Xu, C., Qu, Y., Xiang, Y., and Gao, L. Asynchronous federated learning on heterogeneous devices: A survey. Computer Science Review, 50: 0 100595, 2023

    2023

  56. [64]

    Q., and Chong, K

    Xu, J., Chen, Z., Quek, T. Q., and Chong, K. F. E. Fedcorr: Multi-stage federated learning for label noise correction. In CVPR, pp.\ 10184--10193, 2022

    2022

  57. [65]

    Dynamic personalized federated learning with adaptive differential privacy

    Yang, X., Huang, W., and Ye, M. Dynamic personalized federated learning with adaptive differential privacy. Advances in Neural Information Processing Systems, 36: 0 72181--72192, 2023

    2023

  58. [66]

    Federated graph learning -- a position paper

    Zhang, H., Shen, T., Wu, F., Yin, M., Yang, H., and Wu, C. Federated graph learning -- a position paper. In arXiv preprint arXiv:2105.11099, 2021 a

    2021 arXiv

  59. [67]

    Federated learning with label distribution skew via logits calibration

    Zhang, J., Li, Z., Li, B., Xu, J., Wu, S., Ding, S., and Wu, C. Federated learning with label distribution skew via logits calibration. In ICML, pp.\ 26311--26329, 2022 a

    2022

  60. [68]

    Eliminating domain bias for federated learning in representation space

    Zhang, J., Hua, Y., Cao, J., Wang, H., Song, T., XUE, Z., Ma, R., and Guan, H. Eliminating domain bias for federated learning in representation space. In NeurIPS, pp.\ 14204--14227, 2023 a

    2023

  61. [69]

    Gpfl: Simultaneously learning global and personalized feature information for personalized federated learning

    Zhang, J., Hua, Y., Wang, H., Song, T., Xue, Z., Ma, R., Cao, J., and Guan, H. Gpfl: Simultaneously learning global and personalized feature information for personalized federated learning. In CVPR, pp.\ 5041--5051, 2023 b

    2023

  62. [70]

    Fedala: Adaptive local aggregation for personalized federated learning

    Zhang, J., Hua, Y., Wang, H., Song, T., Xue, Z., Ma, R., and Guan, H. Fedala: Adaptive local aggregation for personalized federated learning. In AAAI, pp.\ 11237--11244, 2023 c

    2023

  63. [71]

    Zhang, K., Yang, C., Li, X., Sun, L., and Yiu, S. M. Subgraph federated learning with missing neighbor generation. In NeurIPS, pp.\ 6671--6682, 2021 b

    2021

  64. [72]

    Timelyfl: Heterogeneity-aware asynchronous federated learning with adaptive partial training

    Zhang, T., Gao, L., Lee, S., Zhang, M., and Avestimehr, S. Timelyfl: Heterogeneity-aware asynchronous federated learning with adaptive partial training. In CVPR, pp.\ 5064--5073, 2023 d

    2023

  65. [73]

    Improving social network embedding via new second-order continuous graph neural networks

    Zhang, Y., Gao, S., Pei, J., and Huang, H. Improving social network embedding via new second-order continuous graph neural networks. In KDD, pp.\ 2515--2523, 2022 b

    2022

  66. [74]

    and Konukoglu, E

    Zhou, T. and Konukoglu, E. Fedfa: Federated feature augmentation. In ICLR, 2023

    2023

  67. [75]

    Zhu, B., Wang, L., Pang, Q., Wang, S., Jiao, J., Song, D., and Jordan, M. I. Byzantine-robust federated learning with optimal statistical rates. In AISTATS, pp.\ 3151--3178, 2023

    2023

  68. [76]

    Fedtad: Topology-aware data-free knowledge distillation for subgraph federated learning

    Zhu, Y., Li, X., Wu, Z., Wu, D., Hu, M., and Li, R.-H. Fedtad: Topology-aware data-free knowledge distillation for subgraph federated learning. arXiv preprint arXiv:2404.14061, 2024

    2024 arXiv

  69. [77]

    Data-free knowledge distillation for heterogeneous federated learning

    Zhu, Z., Hong, J., and Zhou, J. Data-free knowledge distillation for heterogeneous federated learning. In ICML, pp.\ 12878--12889, 2021

    2021

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.