Pith. sign in

REVIEW 4 major objections 3 minor 50 references

Enhancing Federated Graph Learning via Adaptive Fusion of Structural and Node Characteristics

T0 review · 4 major / 3 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Federated graph learning gains 4.9%-7.2% accuracy by fusing structural properties with node features, and cuts communication cost by up to 81%.

desk verdict Test-set feedback in the MAB reward makes the headline accuracy gains unreliable, though the fusion idea is worth a redo. read the letter →

arxiv 2412.18845 v1 pith:7D6VJMCQ submitted 2024-12-25 cs.LG

Xianjun Gao , Jianchun Liu , Hongli Xu , Shilong Wang , Liusheng Huang This is my paper · ORCID
classification cs.LG
keywords federatedgraphlearningnon-IIDdatastructuralpropertiesnodefeaturesadaptivefusionmulti-armedbanditneuralnetworksclientclustering
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 proposes FedGCF, a federated graph learning framework that extracts both structural properties and node features from distributed graph data, then fuses them with an adaptively chosen ratio. The authors claim this combined view outperforms methods that emphasize only one aspect, improving test accuracy by 4.94%-7.24% under different data distributions and reducing communication cost by 64.18%-81.25% to reach the same accuracy as baselines. A sympathetic reader would care because non-IID graph data is common in practice, and existing FGL methods tend to be specialized for either structure-heavy or feature-heavy tasks.

What carries the argument

The key machinery is the combination of Parallel Characteristic Extraction (PCE) and Graph Characteristics Fusion (GCF). PCE encodes structural properties using random-walk position embeddings and maximum-degree embeddings, trains a GIN structural model per client, clusters clients by these structural models, and builds a connected topology from pairwise model similarities (distance $d_{i,j}=e^{\alpha(1-\sigma_{i,j})/(1+\sigma_{i,j})}-1$) to select the top-$P$ longest shortest paths; clients on those paths contribute to the common node model. GCF is a UCB-style multi-armed bandit that chooses the fusion ratio using a reward based on the relative change between current test accuracy $r_t$ and the best historical accuracy $r_b$ (Eq. 6).

What would settle it

Run FedGCF in a setting where the server genuinely has no labeled test data (so Eq. (6) cannot be computed) and compare its accuracy against a fixed-ratio fusion; if the reported 4.94%-7.24% advantage disappears, the claim's load-bearing assumption is falsified.

Watch

Extended reading notes

Core claim

FedGCF is a framework that, in each round, clusters clients by structural similarity and aggregates their structural models within each cluster to form shared structural models, while separately selecting clients with common node features via a connected topology of model similarities and aggregating their local models into a common node model. These two models are then fused with a ratio chosen by a multi-armed bandit algorithm (GCF) that adapts to the training progress and data distribution. The central claim is that this simultaneous extraction and adaptive fusion yields higher accuracy and faster convergence than FedAvg, FedProx, GCFL, and FedStar on benchmark molecule and social-network datasets, including under non-IID skew.

Load-bearing premise

The adaptive fusion ratio relies on the server computing the current test accuracy and comparing it with the best historical accuracy, which assumes the server has access to labeled test data that the privacy-preserving federated setting never explicitly provides.

Editorial extensions

If this is right

  • If FedGCF's claim holds, a single federated framework can handle both structure-heavy tasks (like molecule classification) and feature-heavy tasks (like social-network relationship prediction) without knowing the task type in advance.
  • The adaptive fusion ratio lets the server shift emphasis during training, so the same algorithm works under IID and non-IID data distributions without reconfiguration.
  • The reported communication cost reductions of 64%-81% to reach a target accuracy suggest the method is practical for bandwidth-limited edge deployments.
  • The accuracy gains over FedStar and GCFL imply that structural sharing and feature-based clustering are complementary, and that neither alone captures the information needed for heterogeneous graph data.

Reading between the lines

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

  • A testable extension: applying the same PCE and GCF recipe to heterogeneous graph types (e.g., knowledge graphs) where the optimal emphasis between structure and features also varies per client could show whether the gains generalize beyond the three benchmark families.
  • The reward in Eq. (6) requires the server to compute test accuracy, which implies access to labeled test data; a variant that derives the reward from client-computable statistics (like training loss or validation metrics) would better match the stated privacy-preserving premise.
  • The bandit treats the fusion ratio as a discrete set of actions; a continuous-armed-bandit formulation could smooth the trade-off and possibly converge faster, though the paper's discrete grid is simpler and adequate for the reported experiments.
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

4 major / 3 minor

Summary. The paper proposes FedGCF, a federated graph learning framework that extracts two kinds of graph characteristics separately—structural properties, via structural models clustered by client similarity, and node features, via a client topology and selection of common-feature clients—and then fuses the resulting structural and node models with an adaptively chosen ratio. The ratio is selected by a Multi-Armed Bandit algorithm (GCF). The authors report accuracy gains of 4.94%–7.24% over baselines on Small Molecules, Social Networks, and MIX datasets, and communication-cost reductions of 64.18%–81.25% to reach target accuracy, alongside ablations and scalability experiments.

Significance. The underlying idea—explicitly separating structural and node-feature information and adaptively fusing them under non-IID federated graph distributions—is well motivated and addresses a real limitation of existing FGL methods that emphasize only one type of characteristic. The component-wise ablation in Table IV and Fig. 8 is a useful contribution and suggests the framework's components each matter. However, the central evaluation is compromised by a methodological flaw: the MAB reward in Eq. (6) is computed from test accuracy, and the same test set is used to report the final accuracies in Tables II and III. This makes the headline accuracy and communication-cost claims optimistic and not an unbiased measure of generalization. The paper also does not report variances or specify several important hyperparameters and model-architecture details. If the evaluation is redone with a validation-based reward and a strictly held-out test set, the approach may well show value, but the current manuscript does not establish its central claim.

major comments (4)
  1. [V-B and Eq. (6)] The reward function in Eq. (6) uses r_t, defined as the test accuracy in the current round, and r_b, the best historical test accuracy, and Algorithm 2 uses this reward to select the structural/node-feature fusion ratio. Section V-B states explicitly that "in each round, we will evaluate the global model on the test dataset." Since the final accuracies in Tables II and III are also computed on the test set, the fusion ratio is effectively tuned on the test set, while the baselines receive no such feedback. The reported 4.94%–7.24% gains may therefore be an artifact of test-set feedback rather than of the proposed fusion mechanism. The evaluation must be changed so that the reward is computed on a validation split and the test set is used only once after training is complete; the communication-cost comparisons in Figs. 3–6 should be rerun under the same protocol.
  2. [V-B, Eq. (6), and Algorithm 2] The proposed algorithm assumes the parameter server has access to a labeled test set in every round. This is not stated in the problem formulation and is inconsistent with the paper's privacy-preserving federated learning premise, where clients do not share their local graph data. If a public labeled test set is assumed, the assumption must be stated explicitly and justified; otherwise, Algorithm 2 cannot compute the reward defined in Eq. (6), and the reported accuracies are not achievable in the claimed setting. A validation split available to the server would resolve both this issue and the circularity concern.
  3. [Table II and Section V-C] The paper reports that it averages over "five replicate experiments," but Table II and Table III report only mean accuracy with no standard deviations, confidence intervals, or significance tests. Several comparisons are close (e.g., FedStar at 74.69% vs. FedGCF at 76.38% on Small Molecules), and without variance information it is impossible to judge whether the reported improvements are statistically reliable. Given that the fusion ratio is selected using test-set feedback, this lack of variance reporting makes the headline gains even harder to interpret.
  4. [Section V-A and Fig. 3-6] The communication-cost comparison is not sufficiently specified. FedGCF trains and uploads two models per client (a local model and a structural model), while the baseline architectures are not fully described beyond a shared hidden size of 64. It is unclear whether the reported communication cost for FedGCF includes both uploaded models and whether the baselines use the same architecture (GCN, GIN, or both). Without this information, the claimed communication-cost reduction of 64.18%–81.25% cannot be fairly assessed.
minor comments (3)
  1. [Eq. (7)] The UCB-style score in Eq. (7) is written as sqrt(2 log t) / N_m^t, which differs from the standard UCB form sqrt(2 log t / N_m^t). Please clarify whether this is intentional or a typo, and define the base of the logarithm.
  2. [Section V-B] Several important hyperparameters are not given: the number of client clusters K, the number of selected paths P, the number of candidate combination ratios M, the distance sensitivity alpha in Eq. (5), and the reward amplification factor beta in Eq. (6). These are free parameters that directly affect the reported performance, and their values or selection procedures should be stated for reproducibility.
  3. [Author affiliations] There are spacing artifacts in the author and affiliation lines (e.g., "Liushen g Huang," "University of Sc ience") that should be corrected in the final version.

Circularity Check

1 steps flagged · score 6.0 of 10

Adaptive fusion ratio is selected using test-set accuracy via Eq. (6), and the same test set is then reported as final accuracy, making the headline gains a test-set fit.

  1. fitted input called prediction [Section IV-C, Eq. (6), Algorithm 2; Section V-B]
    "In each round, we will evaluate the global model on the test dataset and record the test accuracy. ... Let r_t represent the test accuracy in the current round t and r_b denote the best test accuracy in history. ... Update the cumulative reward \hat{R}_m for the selected combination ratio m in the last round as Eq. (6)."

    The MAB score \hat{S}_m in Eq. (7) is computed from \hat{R}_m, and Eq. (6) updates \hat{R}_m using r_t and r_b, both of which are test accuracies measured on the same test dataset. Algorithm 2 uses this score to select the structural/node-feature fusion ratio each round. The reported results in Tables II and III are then the test accuracies on that same split. Therefore the adaptive-fusion component is directly tuned to maximize the exact metric reported as the outcome; the claimed gains of 4.94%-7.24% can reflect test-set feedback rather than an unbiased generalization advantage. The baselines are not given the same test-set selection signal, so the comparison is also not controlled.

full rationale

The central adaptive-fusion mechanism is evaluated circularly: the server uses the test dataset to compute the reward that selects the fusion ratio, and the same test dataset is used for the final accuracy tables. This is not a self-citation or uniqueness-theorem issue; it is a protocol in which the evaluation metric is included in the training/selection signal. The structural and node-feature extraction components are not circular and are compared against external baselines, so the paper is not entirely derivative. However, the specific claim that adaptive ratio selection improves accuracy is not supported by a held-out evaluation, because the selected ratio is a function of the test labels at every round. The paper also implicitly assumes the parameter server has access to labeled test data each round, which conflicts with the stated privacy-preserving FGL setting, further confirming that the test-set feedback is load-bearing rather than a harmless evaluation detail. A validation split used only for MAB rewards, with the test set held out until the end, would resolve the circularity; as written, the reported gains are partly a fit to the test set.

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

The framework relies on several free parameters that are not reported, and on domain assumptions about structural encodings and model-similarity proxies. The most serious axiom is the server-side access to test labels, which is necessary for the reward computation but contradicts the stated privacy setting.

free parameters (7)
  • alpha (distance sensitivity in Eq. 5) = not reported
    Controls the exponential mapping from model similarity to client distance; no value or sensitivity analysis is given.
  • beta (reward amplification in Eq. 6) = not reported
    Amplifies reward changes; no value or sensitivity analysis is given.
  • K (number of client clusters) = not reported
    Number of structural clusters in Algorithm 1; no value or selection method is described.
  • P (number of selected longest paths) = not reported
    Number of paths used to choose clients for the common node model; no value is given.
  • M (number of candidate combination ratios) = not reported
    Size of the action set for the MAB; no value or candidate list is provided.
  • reward decay factor and threshold in Eq. 6 = 0.9 and 0.99
    Hand-chosen constants in the reward update; no ablation study is conducted.
  • structural embedding dimensions = 16
    Dimensions for degree and positional embeddings are set to 16 without justification.
assumptions (4)
  • domain assumption Random walk and maximum degree encodings capture graph structural properties sufficiently for clustering and sharing.
    Used in Section IV-B to build structural property vectors su; no theoretical or empirical justification that these encodings capture the needed structure.
  • domain assumption Cosine similarity between client model parameters reflects node-feature similarity between client datasets.
    Used in Eq. 5 to build the client distance topology in Section IV-B; model-parameter similarity is a proxy that may not match data-feature similarity.
  • ad hoc to paper The server has access to a labeled test set to compute rt in the reward function.
    Required by Eq. 6 but absent from the privacy-preserving FGL description; not stated as a public test set or trusted server assumption.
  • domain assumption Clustering structural models groups clients with similar graph topologies.
    Assumed in Algorithm 1, Lines 1-2; no evaluation of clustering quality or stability is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing Federated Graph Learning via Adaptive Fusion of Structural and Node Characteristics." pith.science (2026). https://pith.science/paper/7D6VJMCQ

@misc{pith2026241218845,
  author       = {Pith},
  title        = {Pith review of: Enhancing Federated Graph Learning via Adaptive Fusion of Structural and Node Characteristics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7D6VJMCQ}},
  note         = {Machine review of arXiv:2412.18845}
}
read the original abstract

Federated Graph Learning (FGL) has demonstrated the advantage of training a global Graph Neural Network (GNN) model across distributed clients using their local graph data. Unlike Euclidean data (\eg, images), graph data is composed of nodes and edges, where the overall node-edge connections determine the topological structure, and individual nodes along with their neighbors capture local node features. However, existing studies tend to prioritize one aspect over the other, leading to an incomplete understanding of the data and the potential misidentification of key characteristics across varying graph scenarios. Additionally, the non-independent and identically distributed (non-IID) nature of graph data makes the extraction of these two data characteristics even more challenging. To address the above issues, we propose a novel FGL framework, named FedGCF, which aims to simultaneously extract and fuse structural properties and node features to effectively handle diverse graph scenarios. FedGCF first clusters clients by structural similarity, performing model aggregation within each cluster to form the shared structural model. Next, FedGCF selects the clients with common node features and aggregates their models to generate a common node model. This model is then propagated to all clients, allowing common node features to be shared. By combining these two models with a proper ratio, FedGCF can achieve a comprehensive understanding of the graph data and deliver better performance, even under non-IID distributions. Experimental results show that FedGCF improves accuracy by 4.94%-7.24% under different data distributions and reduces communication cost by 64.18%-81.25% to reach the same accuracy compared to baselines.

Figures

Figures reproduced from arXiv: 2412.18845 by the authors.

Figure 1
Figure 1. The impact of different data characteristics on two d [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the Graph Characteristics Extracti [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. The test accuracy and communication cost for FedGCF a [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (4 more)
Figure 5
Figure 5. Figure 5: The test accuracy and communication cost for FedGCF [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Communication cost to reach different target model t [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 9
Figure 9. Figure 9: Communication cost to reach different target model t [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 8
Figure 8. Figure 8: The model test accuracy of FedGCF and three segmented [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 35 canonical work pages

  1. [1]

    Graph convolutional neural networks for web- scale rec- ommender systems,

    R. Ying, R. He, K. Chen, P . Eksombatchai, W. L. Hamilton, a nd J. Leskovec, “Graph convolutional neural networks for web- scale rec- ommender systems,” in Proceedings of the 24th ACM SIGKDD inter- national conference on knowledge discovery & data mining , 2018, pp. 974–983

    work page 2018

  2. [2]

    Measurement and analysis of online social networks,

    A. Mislove, M. Marcon, K. P . Gummadi, P . Druschel, and B. B hattachar- jee, “Measurement and analysis of online social networks,” in Proceed- ings of the 7th ACM SIGCOMM conference on Internet measureme nt, 2007, pp. 29–42

    work page 2007

  3. [3]

    Representatio n learning on graphs: Methods and applications,

    W. L. Hamilton, R. Ying, and J. Leskovec, “Representatio n learning on graphs: Methods and applications,” arXiv preprint arXiv:1709.05584 , 2017

    arXiv 2017

  4. [4]

    He teroge- neous graph neural network,

    C. Zhang, D. Song, C. Huang, A. Swami, and N. V . Chawla, “He teroge- neous graph neural network,” in Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data min ing, 2019, pp. 793–803

    work page 2019

  5. [5]

    Simplifying graph convolutional networks,

    F. Wu, A. Souza, T. Zhang, C. Fifty, T. Y u, and K. Weinberge r, “Simplifying graph convolutional networks,” in International conference on machine learning . PMLR, 2019, pp. 6861–6871

    work page 2019

  6. [6]

    Deepgcns: Can gcns go as deep as cnns?

    G. Li, M. Muller, A. Thabet, and B. Ghanem, “Deepgcns: Can gcns go as deep as cnns?” in Proceedings of the IEEE/CVF international conference on computer vision , 2019, pp. 9267–9276

    2019

  7. [7]

    Federated lea rning-based cross-enterprise recommendation with graph neural networ ks,

    Z. Li, M. Bilal, X. Xu, J. Jiang, and Y . Cui, “Federated lea rning-based cross-enterprise recommendation with graph neural networ ks,” IEEE Transactions on Industrial Informatics, vol. 19, no. 1, pp. 673–682, 2022

    work page 2022

  8. [8]

    Fedgcn: Converg ence- communication tradeoffs in federated training of graph con volutional networks,

    Y . Y ao, W. Jin, S. Ravi, and C. Joe-Wong, “Fedgcn: Converg ence- communication tradeoffs in federated training of graph con volutional networks,” Advances in neural information processing systems , vol. 36, 2024

    work page 2024

Show all 50 references
  1. [9]

    Glasu: A communication-e fficient algorithm for federated learning with vertically distribu ted graph data,

    X. Zhang, M. Hong, and J. Chen, “Glasu: A communication-e fficient algorithm for federated learning with vertically distribu ted graph data,” arXiv preprint arXiv:2303.09531 , 2023

    2023 arXiv

  2. [10]

    Fedgnn: Federa ted graph neural network for privacy-preserving recommendation,

    C. Wu, F. Wu, Y . Cao, Y . Huang, and X. Xie, “Fedgnn: Federa ted graph neural network for privacy-preserving recommendation,” arXiv preprint arXiv:2102.04925, 2021

    2021 arXiv

  3. [11]

    Enhan cing federated learning with in-cloud unlabeled data,

    L. Wang, Y . Xu, H. Xu, J. Liu, Z. Wang, and L. Huang, “Enhan cing federated learning with in-cloud unlabeled data,” in 2022 IEEE 38th International Conference on Data Engineering (ICDE) . IEEE, 2022, pp. 136–149

    2022

  4. [12]

    Fedm p: Federated learning through adaptive model pruning in heter ogeneous edge computing,

    Z. Jiang, Y . Xu, H. Xu, Z. Wang, C. Qiao, and Y . Zhao, “Fedm p: Federated learning through adaptive model pruning in heter ogeneous edge computing,” in 2022 IEEE 38th International Conference on Data Engineering (ICDE) . IEEE, 2022, pp. 767–779

    2022

  5. [13]

    Federated learni ng for vision-and-language grounding problems,

    F. Liu, X. Wu, S. Ge, W. Fan, and Y . Zou, “Federated learni ng for vision-and-language grounding problems,” in Proceedings of the AAAI conference on artificial intelligence , vol. 34, no. 07, 2020, pp. 11 572– 11 579

    2020

  6. [14]

    Peaches: Pers onalized federated learning with neural architecture search in edge computing,

    J. Y an, J. Liu, H. Xu, Z. Wang, and C. Qiao, “Peaches: Pers onalized federated learning with neural architecture search in edge computing,” IEEE Transactions on Mobile Computing , 2024

    2024

  7. [15]

    The structure and function of complex net works,

    M. E. Newman, “The structure and function of complex net works,” SIAM review, vol. 45, no. 2, pp. 167–256, 2003

    2003

  8. [16]

    Clayden, N

    J. Clayden, N. Greeves, and S. Warren, Organic chemistry . Oxford University Press, USA, 2012

    2012

  9. [17]

    Social network analysis: An in troduction,

    A. Marin and B. Wellman, “Social network analysis: An in troduction,” The SAGE handbook of social network analysis , pp. 11–25, 2011

    2011

  10. [18]

    Towards communi cation- efficient federated graph learning: An adaptive client sele ction perspec- tive,

    X. Gao, J. Liu, H. Xu, Q. Ma, and L. Wang, “Towards communi cation- efficient federated graph learning: An adaptive client sele ction perspec- tive,” in 2024 IEEE/ACM 32nd International Symposium on Quality of Service (IWQoS) . IEEE, 2024, pp. 1–10

    2024

  11. [19]

    Federated learning with non-iid data,

    Y . Zhao, M. Li, L. Lai, N. Suda, D. Civin, and V . Chandra, “ Federated learning with non-iid data,” arXiv preprint arXiv:1806.00582 , 2018

    2018 arXiv

  12. [20]

    Non-iid transfer learni ng on graphs,

    J. Wu, J. He, and E. Ainsworth, “Non-iid transfer learni ng on graphs,” in Proceedings of the AAAI Conference on Artificial Intelligen ce, vol. 37, no. 9, 2023, pp. 10 342–10 350

    2023

  13. [21]

    Fine- tuning global model via data-free knowledge distillation for non- iid federated learning,

    L. Zhang, L. Shen, L. Ding, D. Tao, and L.-Y . Duan, “Fine- tuning global model via data-free knowledge distillation for non- iid federated learning,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2022, pp. 10 174–10 183

    2022

  14. [22]

    On the con vergence of fedavg on non-iid data,

    X. Li, K. Huang, W. Y ang, S. Wang, and Z. Zhang, “On the con vergence of fedavg on non-iid data,” arXiv preprint arXiv:1907.02189 , 2019

    1907 arXiv

  15. [23]

    Federated graph cla ssification over non-iid graphs,

    H. Xie, J. Ma, L. Xiong, and C. Y ang, “Federated graph cla ssification over non-iid graphs,” Advances in neural information processing systems, vol. 34, pp. 18 839–18 852, 2021

    2021

  16. [24]

    Fed erated learn- ing on non-iid graphs via structural knowledge sharing,

    Y . Tan, Y . Liu, G. Long, J. Jiang, Q. Lu, and C. Zhang, “Fed erated learn- ing on non-iid graphs via structural knowledge sharing,” in Proceedings of the AAAI conference on artificial intelligence , vol. 37, no. 8, 2023, pp. 9953–9961

    2023

  17. [25]

    Federated graph seman tic and structural learning,

    W. Huang, G. Wan, M. Y e, and B. Du, “Federated graph seman tic and structural learning,” arXiv preprint arXiv:2406.18937 , 2024

    2024 arXiv

  18. [26]

    Fedgraphnn: A federated learning benchmark system for graph neural networks,

    C. He, K. Balasubramanian, E. Ceyani, C. Y ang, H. Xie, L. Sun, L. He, L. Y ang, S. Y . Philip, Y . Ronget al., “Fedgraphnn: A federated learning benchmark system for graph neural networks,” in ICLR 2021 W orkshop on Distributed and Private Machine Learning (DPML) , 2021

    2021

  19. [27]

    Semi-supervised classificat ion with graph convolutional networks,

    T. N. Kipf and M. Welling, “Semi-supervised classificat ion with graph convolutional networks,” arXiv preprint arXiv:1609.02907 , 2016

    2016 arXiv

  20. [28]

    Inductive repre sentation learn- ing on large graphs,

    W. Hamilton, Z. Ying, and J. Leskovec, “Inductive repre sentation learn- ing on large graphs,” Advances in neural information processing systems , vol. 30, 2017

    2017

  21. [29]

    Communication-efficient learning of deep networks from de centralized data,

    B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Ar cas, “Communication-efficient learning of deep networks from de centralized data,” in Artificial intelligence and statistics . PMLR, 2017, pp. 1273– 1282

    2017

  22. [30]

    Ligra: a lightweight graph p rocessing frame- work for shared memory,

    J. Shun and G. E. Blelloch, “Ligra: a lightweight graph p rocessing frame- work for shared memory,” in Proceedings of the 18th ACM SIGPLAN symposium on Principles and practice of parallel programmi ng, 2013, pp. 135–146

    2013

  23. [31]

    Optimizing graph algorithm s on pregel-like systems,

    S. Salihoglu and J. Widom, “Optimizing graph algorithm s on pregel-like systems,” 2014

    2014

  24. [32]

    Scalable gpu g raph traver- sal,

    D. Merrill, M. Garland, and A. Grimshaw, “Scalable gpu g raph traver- sal,” ACM Sigplan Notices , vol. 47, no. 8, pp. 117–128, 2012

    2012

  25. [33]

    Neural message passing for quantum chemistry,

    J. Gilmer, S. S. Schoenholz, P . F. Riley, O. Vinyals, and G. E. Dahl, “Neural message passing for quantum chemistry,” in International con- ference on machine learning . PMLR, 2017, pp. 1263–1272

    2017

  26. [34]

    Graph neural networks: A review of methods and appli cations,

    J. Zhou, G. Cui, S. Hu, Z. Zhang, C. Y ang, Z. Liu, L. Wang, C . Li, and M. Sun, “Graph neural networks: A review of methods and appli cations,” AI open , vol. 1, pp. 57–81, 2020

    2020

  27. [35]

    The non- iid data quagmire of decentralized machine learning,

    K. Hsieh, A. Phanishayee, O. Mutlu, and P . Gibbons, “The non- iid data quagmire of decentralized machine learning,” in International Conference on Machine Learning . PMLR, 2020, pp. 4387–4398

    2020

  28. [36]

    Netgan: Generating graphs via random walks,

    A. Bojchevski, O. Shchur, D. Zügner, and S. Günnemann, “ Netgan: Generating graphs via random walks,” in International conference on machine learning. PMLR, 2018, pp. 610–619

    2018

  29. [37]

    Adaptive federated optimizati on,

    S. Reddi, Z. Charles, M. Zaheer, Z. Garrett, K. Rush, J. K oneˇcn`y, S. Kumar, and H. B. McMahan, “Adaptive federated optimizati on,” arXiv preprint arXiv:2003.00295, 2020

    2003 arXiv

  30. [38]

    Random w alks: A review of algorithms and applications,

    F. Xia, J. Liu, H. Nie, Y . Fu, L. Wan, and X. Kong, “Random w alks: A review of algorithms and applications,” IEEE Transactions on Emerging Topics in Computational Intelligence , vol. 4, no. 2, pp. 95–107, 2019

    2019

  31. [39]

    Fast random walk w ith restart and its applications,

    H. Tong, C. Faloutsos, and J.-Y . Pan, “Fast random walk w ith restart and its applications,” in Sixth international conference on data mining (ICDM’06). IEEE, 2006, pp. 613–622

    2006

  32. [40]

    On the vertex-degree-function indices of connected (n, m) -graphs of maximum degree at most four,

    A. M. Albalahi, I. Z. Milovanovic, Z. Raza, A. Ali, and A. E. Hamza, “On the vertex-degree-function indices of connected (n, m) -graphs of maximum degree at most four,” arXiv preprint arXiv:2207.00353 , 2022

    2022 arXiv

  33. [41]

    Unifying pac a nd regret: Uniform pac bounds for episodic reinforcement learning,

    C. Dann, T. Lattimore, and E. Brunskill, “Unifying pac a nd regret: Uniform pac bounds for episodic reinforcement learning,” Advances in Neural Information Processing Systems , vol. 30, 2017

    2017

  34. [42]

    Trust region policy optimization,

    J. Schulman, “Trust region policy optimization,” arXiv preprint arXiv:1502.05477, 2015

    2015 arXiv

  35. [43]

    Prox- imal policy optimization algorithms,

    J. Schulman, F. Wolski, P . Dhariwal, A. Radford, and O. K limov, “Prox- imal policy optimization algorithms,” arXiv preprint arXiv:1707.06347 , 2017

    2017 arXiv

  36. [44]

    Finite-time analysis of the multiarmed bandi t problem,

    P . Auer, “Finite-time analysis of the multiarmed bandi t problem,” 2002

    2002

  37. [45]

    A survey on contextual multi-armed bandits,

    L. Zhou, “A survey on contextual multi-armed bandits,” arXiv preprint arXiv:1508.03326, 2015

    2015 arXiv

  38. [46]

    A context ual-bandit approach to personalized news article recommendation,

    L. Li, W. Chu, J. Langford, and R. E. Schapire, “A context ual-bandit approach to personalized news article recommendation,” in Proceedings of the 19th international conference on W orld wide web , 2010, pp. 661– 670

    2010

  39. [47]

    How powerful a re graph neural networks?

    K. Xu, W. Hu, J. Leskovec, and S. Jegelka, “How powerful a re graph neural networks?” arXiv preprint arXiv:1810.00826 , 2018

    2018 arXiv

  40. [48]

    Federated optimization in heterogeneous networks,

    T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, an d V . Smith, “Federated optimization in heterogeneous networks,” Proceedings of Machine learning and systems , vol. 2, pp. 429–450, 2020

    2020

  41. [49]

    Adam: A method for stochastic optimizati on,

    D. P . Kingma, “Adam: A method for stochastic optimizati on,” arXiv preprint arXiv:1412.6980, 2014

    2014 arXiv

  42. [50]

    On the convergence of adam and beyond,

    S. J. Reddi, S. Kale, and S. Kumar, “On the convergence of adam and beyond,” arXiv preprint arXiv:1904.09237 , 2019. 13 This figure "fig1.png" is available in "png" format from: http://arxiv.org/ps/2412.18845v1

    1904 arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.