Pith. sign in

REVIEW 3 major objections 5 minor 102 references

Role-aware clustering compresses multi-type graphs into small training graphs that keep HGNN accuracy high without expensive optimization.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-12 04:55 UTC pith:5GJSSDIZ

load-bearing objection Optimization-free heterogeneous condensation via role-aware clustering is a real, practical recipe; the main open question is backbone transfer, not the core idea. the 3 major comments →

arxiv 2607.03097 v1 pith:5GJSSDIZ submitted 2026-07-03 cs.LG

Heterogeneous Graph Condensation via Role-Aware Clustering

Fuyan Ou , Yulin Hu , Ye Yuan This is my paper
classification cs.LG
keywords heterogeneous graph condensationheterogeneous graph neural networksrole-aware clusteringgraph reductionmeta-path embeddingsnode classification
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Training heterogeneous graph neural networks on large multi-type graphs is costly because of relation-aware message passing and high-dimensional features. This paper argues that you do not need gradient matching or bilevel optimization to compress such graphs. Instead, first build semantic node embeddings by light propagation, then cluster labeled target nodes separately by class so class balance is kept, and cluster other node types unsupervised so support structure is kept. Super-nodes are formed by averaging features inside each cluster, and edges are rebuilt from normalized, thresholded inter-cluster counts. On three standard benchmarks the resulting tiny graphs let an HGNN match or beat prior condensation methods under high compression while finishing condensation much faster.

Core claim

A compact heterogeneous training graph built by class-partitioned clustering of labeled target nodes plus type-wise unsupervised clustering of non-target nodes, followed by mean-feature aggregation and thresholded inter-cluster adjacency, preserves enough downstream node-classification utility that an HGNN trained on it reaches competitive or best accuracy under severe size reduction, without iterative synthetic-graph optimization.

What carries the argument

Role-aware hybrid clustering: labeled target nodes are clustered inside each class to keep decision-boundary structure; non-target nodes are clustered by type without labels to keep cross-type support; the resulting clusters become the condensed nodes whose features and edges are reconstructed in closed form.

Load-bearing premise

The method assumes that SeHGNN-style embeddings plus this hybrid clustering and simple edge-threshold reconstruction still carry enough of the original multi-relation training signal that accuracy on the tiny graph is a fair stand-in for full-graph utility.

What would settle it

Train the same HGNN backbone on HGC-RC condensed graphs versus full graphs (and versus HGCond) on a held-out large heterogeneous benchmark; if accuracy collapses relative to the full graph or falls well below HGCond under the same budget while condensation time remains low, the claim fails.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes HGC-RC, an optimization-free framework for heterogeneous graph condensation. It first builds semantic embeddings with SeHGNN-style meta-path propagation, then applies role-aware hybrid clustering: class-partitioned clustering on labeled target nodes to preserve class balance, and unsupervised type-wise clustering on non-target nodes to retain support structure. Condensed features are mean-aggregated within clusters; multi-relation adjacencies are reconstructed from normalized inter-cluster edge counts, sigmoid scoring, and thresholding. On ACM, DBLP, and IMDB under ratios 1.2%–9.6%, HGC-RC is competitive with or better than Random/K-Center/Coarsening-HG, GCond, GCPA, and HGCond on SeHGNN node classification, with substantially lower condensation time than optimization-based methods. Ablations remove semantic embedding and class partition.

Significance. If the results hold more broadly, the work offers a practical, low-cost alternative to gradient-matching/bilevel heterogeneous condensation. The role asymmetry between target and non-target nodes is a clear design principle, and the one-shot clustering-plus-reconstruction pipeline is easy to implement and much faster than HGCond/GCond (Fig. 3). Strengths include a coherent empirical package (three HGB-style datasets, multiple ratios, reduction and condensation baselines, ablations in Table III, runtime comparison) and an explicit, reproducible reconstruction rule. The main limitation on significance is that all utility claims currently rest on a single matching backbone (SeHGNN for both embedding extraction and downstream evaluation), so the claimed “practical pathway to accelerate HGNN training” is only demonstrated for that family.

major comments (3)
  1. All main results (Table I), ablations (Table III), and the efficiency claim (Fig. 3) use SeHGNN both to produce the embeddings for clustering (§III.A) and as the sole downstream HGNN. The abstract and §V claim a pathway to accelerate HGNN training in general, but there is no transfer experiment to other HGNNs (e.g., HAN, HGT, MAGNN, or a simple relation-aware GCN). Without at least one non-SeHGNN backbone trained on the same condensed graphs, it remains open whether the clusters and reconstructed multi-relation structure preserve model-agnostic training utility or mainly SeHGNN-specific meta-path geometry. This is load-bearing for the central claim and should be addressed with additional evaluation or a clearly scoped claim.
  2. Table I shows HGC-RC underperforms HGCond on DBLP at 1.2% and 2.4% (and is only slightly better at higher ratios), while dominating on ACM/IMDB. The text (§IV.B) notes the pattern but does not analyze when role-aware clustering fails under tight budgets (e.g., class-partitioned k_c becoming 1, loss of cross-type support, sensitivity of Ac_r thresholds). A short failure analysis or budget-sensitivity study on DBLP would strengthen the claim that the method is robust under high compression rather than dataset-dependent.
  3. Homogeneous baselines GCond and GCPA are transferred via a “homogenized graph view” (§IV.A.2) without specifying how multi-type features and multi-relation edges are collapsed, or whether that homogenization is fair relative to HGCond/HGC-RC which keep typed structure. Because these methods often trail substantially in Table I, the comparison needs an explicit homogenization protocol (and ideally a brief check that the same protocol does not artificially harm them) so that the SOTA claim is interpretable.
minor comments (5)
  1. Table I has duplicated HGC-RC columns and repeated accuracy strings (e.g., ACM 1.2% row); clean the layout so each method appears once.
  2. §III.C: α=10, δ=0.5, and β_r = mean of nonzero normalized scores are fixed without sensitivity analysis; a short note or appendix on robustness would help reproducibility.
  3. Fig. 1 caption and panel labels are schematic; a one-sentence concrete example (e.g., paper vs. author/term on ACM) would make the role-asymmetry motivation clearer.
  4. Notation: condensation mask M is introduced in §II.A and used in §III.B; state explicitly that validation/test targets are never condensed and remain on the original graph for evaluation (implied but easy to miss).
  5. Related work could briefly position Graph-Skeleton (already cited) against the full target+non-target compression setting to avoid reader confusion about problem scope.

Circularity Check

0 steps flagged

No significant circularity: empirical condensation method whose accuracy claims are measured on held-out data, not forced by construction or self-citation.

full rationale

HGC-RC is a data-centric engineering pipeline (SeHGNN embedding extraction, class-partitioned/type-wise clustering, mean-feature + thresholded inter-cluster adjacency reconstruction) whose utility is assessed by training an HGNN on the resulting compact graph and reporting node-classification accuracy against external baselines and the full graph (Table I, Fig. 3, ablations). Class balance is deliberately preserved by construction in the target-node budget allocation, but the paper never presents the measured test accuracy as a derived or predicted quantity; it is an empirical outcome under a fixed evaluation protocol. Hyper-parameters (α=10, δ=0.5, β_r = mean of nonzero scores, SeHGNN dimensions/layers) are author-chosen constants, not fitted quantities later re-labeled as predictions. Background citations (GCond, HGCond, SeHGNN, etc.) supply prior art rather than load-bearing uniqueness theorems authored by the present team. The matching of SeHGNN for both embedding and downstream evaluation is a scope limitation, not a circular reduction of the claim to its inputs. Consequently the derivation chain contains no self-definitional step, fitted-input-as-prediction, or self-citation that forces the reported results.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 2 invented entities

The central empirical claim rests on standard graph-learning practice plus a few paper-specific design axioms and hand-set reconstruction parameters. No new physical entities; the invented pieces are methodological constructs (role-aware hybrid clustering, the condensed multi-relation reconstruction rule). Free parameters are the usual ML knobs plus three adjacency thresholds that directly shape Gc.

free parameters (5)
  • adjacency sharpness α = 10
    Fixed to 10 in §III.C / §IV.A.3; controls sigmoid steepness of inter-cluster edge scores and thus which condensed edges exist.
  • edge filter threshold δ = 0.5
    Fixed to 0.5; binarizes Sc_r into Ac_r and directly determines condensed topology density.
  • relation threshold β_r = mean of nonzero fW^c_r
    Set as mean of nonzero normalized inter-cluster counts per relation; shifts the sigmoid and is data-dependent without cross-validation reported.
  • condensation ratio ρ and type/class budgets = 1.2–9.6% (evaluated grid)
    ρ ∈ {1.2%, 2.4%, 4.8%, 9.6%} with type-wise and class-proportional integer budgets; defines the size of Gc the claims are measured under.
  • SeHGNN embedding / training hyperparameters = dim=512; dataset-specific layers
    Hidden/embedding dim 512, 2 prep layers, dataset-specific task layers, dropout 0.5, Adam lr 0.001, 200 epochs, patience 50; these define the clustering space Z.
axioms (5)
  • domain assumption Labeled target-type nodes and unlabeled non-target nodes play asymmetric roles for downstream classification utility, so they should be condensed with different strategies.
    Stated as the key challenge in §I and Fig. 1; load-bearing for the hybrid clustering design.
  • domain assumption SeHGNN meta-path preprocessing yields an embedding space in which clustering preserves both class structure and cross-type support.
    §III.A instantiates g(·) as SeHGNN; ablations in Table III show large drops without it, so the claim depends on this representation choice.
  • ad hoc to paper Mean feature aggregation over clusters and thresholded normalized inter-cluster edge counts sufficiently reconstruct a training graph for HGNNs.
    §III.C defines xc_k, W/fW, Sc, Ac; no theoretical guarantee that this matches original training dynamics.
  • domain assumption Condensing only the labeled training mask M of target nodes, while evaluating on original validation/test nodes, is a valid condensation protocol.
    §II.A explicitly adopts this train-graph condensation / original-graph evaluation split common in GC.
  • domain assumption Homogenized transfers of GCond/GCPA are fair heterogeneous baselines when compared under the same ρ.
    §IV.A.2; fairness of homogenization is assumed rather than proven.
invented entities (2)
  • HGC-RC role-aware hybrid clustering no independent evidence
    purpose: Map original multi-type nodes to a small set of condensed nodes while preserving class balance for targets and support connectivity for non-targets.
    Core methodological construct of the paper (§III.B); not an external physical entity. Independent evidence is only the paper’s own accuracy/runtime tables.
  • Cluster-level multi-relation reconstruction rule (normalized counts → sigmoid scores → thresholded Ac_r) no independent evidence
    purpose: Build condensed heterogeneous adjacencies without gradient matching.
    Defined in §III.C with paper-chosen α/δ/β_r; success is measured only by downstream accuracy in this work.

pith-pipeline@v1.1.0-grok45 · 19460 in / 3965 out tokens · 40564 ms · 2026-07-12T04:55:02.537114+00:00 · methodology

0 comments
read the original abstract

Heterogeneous Graph Neural Networks (HGNNs) have exhibited remarkable efficacy in modeling complex systems with multiple types of nodes and relations, yet their training on large-scale heterogeneous graphs remains computationally prohibitive. Although graph condensation methods can effectively improve learning efficiency on large-scale graphs, existing condensation processes are mainly designed for homogeneous graphs and typically rely on computationally expensive gradient matching or bilevel optimization paradigms, rendering them impractical for heterogeneous settings. To address these limitations, we propose HGC-RC, a simple yet effective role-aware heterogeneous graph condensation framework. Specifically, HGC-RC first extracts semantically enhanced node embeddings via lightweight propagation. It then introduces a role-aware hybrid clustering strategy consisting of class-partitioned clustering for labeled target nodes to preserve class distributions and unsupervised type-wise clustering for non-target nodes to retain critical cross-type connectivity. Finally, a compact heterogeneous graph is efficiently reconstructed based on the resulting cluster assignments. Extensive experiments demonstrate that HGC-RC outperforms state-of-the-art baselines, offering a practical pathway to accelerate HGNN training on large-scale heterogeneous graphs without sacrificing task performance

Figures

Figures reproduced from arXiv: 2607.03097 by Fuyan Ou, Ye Yuan, Yulin Hu.

Figure 1
Figure 1. Figure 1: Motivation of role-aware condensation in heterogeneous graphs. (a) Uniform condensation suits homogeneous graphs. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The workflow of HGC-RC. Let Z ∈ R |V|×d denote the embedding matrix, where the row vector zv corresponds to node v. We write the embedding extraction stage as Z = g  G, {XP }P ∈P  , (7) where g(·) is instantiated by a SeHGNN [102] preprocessing encoder to capture multi-relational semantics with low overhead. The resulting embeddings encode both structural context and semantic patterns induced by meta-pat… view at source ↗
Figure 3
Figure 3. Figure 3: Runtime comparison with other methods. 2) Baselines: We compare HGC-RC with representative graph reduction and condensation baselines, including Random￾HG, K-Center-HG, Coarsening-HG, GCond, GCPA, and HGCond. Random-HG randomly samples nodes under the same budget. K-Center-HG selects representative nodes in the embedding space. Coarsening-HG applies topology-based graph coarsening. GCond and GCPA are homog… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

102 extracted references · 7 linked inside Pith

  1. [1]

    A generalized and fast-converging non-negative latent factor model for predicting user preferences in recommender systems,

    Y . Yuan, X. Luo, M. Shang, and D. Wu, “A generalized and fast-converging non-negative latent factor model for predicting user preferences in recommender systems,” inThe Web Conference, 2020, pp. 498–507

    2020

  2. [2]

    Auto-encoding neural tucker factorization,

    P. Tang, X. Luo, and J. Woodcock, “Auto-encoding neural tucker factorization,”IEEE Transactions on Knowledge and Data Engineering, vol. 37, no. 10, pp. 5795–5807, 2025

    2025

  3. [3]

    A proximal-admm-incorporated nonnegative latent-factorization-of-tensors model for representing dynamic cryptocurrency transaction network,

    X. Liao, H. Wu, T. He, and X. Luo, “A proximal-admm-incorporated nonnegative latent-factorization-of-tensors model for representing dynamic cryptocurrency transaction network,”IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 55, no. 11, pp. 8387–8401, 2025

    2025

  4. [4]

    A comprehensive review of parallel optimization algorithms for high-dimensional and incomplete matrix factorization,

    Q. Hu, H. Wu, and X. Luo, “A comprehensive review of parallel optimization algorithms for high-dimensional and incomplete matrix factorization,” IEEE/CAA Journal of Automatica Sinica, vol. 12, no. 12, pp. 2399–2426, 2025

    2025

  5. [5]

    Multi-scale collaborative distillation graph neural networks for session-based recommendation,

    J. Gou, Y . Cheng, B. Ma, L. Du, X. Luo, and Z. Yi, “Multi-scale collaborative distillation graph neural networks for session-based recommendation,” IEEE Transactions on Services Computing, vol. 19, no. 1, pp. 504–517, 2026

    2026

  6. [6]

    Adaptively-accelerated parallel stochastic gradient descent for high-dimensional and incomplete data representation learning,

    W. Qin, X. Luo, and M. Zhou, “Adaptively-accelerated parallel stochastic gradient descent for high-dimensional and incomplete data representation learning,”IEEE Transactions on Big Data, vol. 10, no. 1, pp. 92–107, 2024

    2024

  7. [7]

    Semi-supervised classification with graph convolutional networks,

    T. N. Kipf and M. Welling, “Semi-supervised classification with graph convolutional networks,” in5th International Conference on Learning Representations, ICLR 2017, 2017

    2017

  8. [8]

    Graph attention networks,

    P. Velickovic, G. Cucurull, A. Casanova, A. Romero, P. Li `o, and Y . Bengio, “Graph attention networks,” in6th International Conference on Learning Representations, ICLR 2018, 2018

    2018

  9. [9]

    Inductive representation learning on large graphs,

    W. L. Hamilton, Z. Ying, and J. Leskovec, “Inductive representation learning on large graphs,” inAdvances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, 2017, pp. 1024–1034

    2017

  10. [10]

    Fastgcn: Fast learning with graph convolutional networks via importance sampling,

    J. Chen, T. Ma, and C. Xiao, “Fastgcn: Fast learning with graph convolutional networks via importance sampling,” in6th International Conference on Learning Representations, ICLR 2018, 2018

    2018

  11. [11]

    Graphsaint: Graph sampling based inductive learning method,

    H. Zeng, H. Zhou, A. Srivastava, R. Kannan, and V . K. Prasanna, “Graphsaint: Graph sampling based inductive learning method,” in8th International Conference on Learning Representations, ICLR, 2020

    2020

  12. [12]

    A node-collaboration-informed graph convolutional network for highly accurate representation to undirected weighted graph,

    Y . Yuan, Y . Wang, and X. Luo, “A node-collaboration-informed graph convolutional network for highly accurate representation to undirected weighted graph,”IEEE Transactions on Neural Networks and Learning Systems, vol. 36, no. 6, pp. 11 507–11 519, 2025

    2025

  13. [13]

    Modularized graph convolutional network,

    T. He, Z. Duan, and X. Luo, “Modularized graph convolutional network,”IEEE/CAA Journal of Automatica Sinica, 2025

    2025

  14. [14]

    Graph linear convolution pooling for learning in incomplete high-dimensional data,

    F. Bi, T. He, Y .-S. Ong, and X. Luo, “Graph linear convolution pooling for learning in incomplete high-dimensional data,”IEEE Transactions on Knowledge and Data Engineering, vol. 37, no. 4, pp. 1838–1852, 2025

    2025

  15. [15]

    A generalized nesterov-accelerated second-order latent factor model for high-dimensional and incomplete data,

    W. Li, R. Wang, and X. Luo, “A generalized nesterov-accelerated second-order latent factor model for high-dimensional and incomplete data,”IEEE Transactions on Neural Networks and Learning Systems, vol. 36, no. 1, pp. 1518–1532, 2024

    2024

  16. [16]

    Graph tensor convolutional network,

    L. Wang, Y . Yuan, and X. Luo, “Graph tensor convolutional network,”IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2026

    2026

  17. [17]

    Advanced high-order graph convolutional networks with assorted time-frequency transforms,

    ——, “Advanced high-order graph convolutional networks with assorted time-frequency transforms,”IEEE/CAA Journal of Automatica Sinica, vol. 13, no. 2, pp. 394–408, 2026

    2026

  18. [18]

    Tensor low-rank orthogonal compression for convolutional neural networks,

    Y . He and X. Luo, “Tensor low-rank orthogonal compression for convolutional neural networks,”IEEE/CAA Journal of Automatica Sinica, vol. 13, no. 1, pp. 227–229, 2026

    2026

  19. [19]

    Heterogeneous graph attention network,

    X. Wang, H. Ji, C. Shi, B. Wang, Y . Ye, P. Cui, and P. S. Yu, “Heterogeneous graph attention network,” inThe World Wide Web Conference, WWW 2019, 2019, pp. 2022–2032

    2019

  20. [20]

    Heterogeneous graph neural network,

    C. Zhang, D. Song, C. Huang, A. Swami, and N. V . Chawla, “Heterogeneous graph neural network,” inProceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 2019

    2019

  21. [21]

    MAGNN: metapath aggregated graph neural network for heterogeneous graph embedding,

    X. Fu, J. Zhang, Z. Meng, and I. King, “MAGNN: metapath aggregated graph neural network for heterogeneous graph embedding,” inWWW ’20: The Web Conference 2020, 2020, pp. 2331–2341

    2020

  22. [22]

    An attention-based graph neural network for heterogeneous structural learning,

    H. Hong, H. Guo, Y . Lin, X. Yang, Z. Li, and J. Ye, “An attention-based graph neural network for heterogeneous structural learning,” inThe Thirty-Fourth AAAI Conference on Artificial Intelligence, AAAI 2020, 2020, pp. 4132–4139

    2020

  23. [23]

    Scalable graph neural networks for heterogeneous graphs,

    L. Yu, J. Shen, J. Li, and A. Lerer, “Scalable graph neural networks for heterogeneous graphs,”arXiv preprint arXiv:2011.09679, 2020

    Pith/arXiv arXiv 2011

  24. [24]

    Gt-a2t: Graph tensor alliance attention network,

    L. Wang, K. Liu, and Y . Yuan, “Gt-a2t: Graph tensor alliance attention network,”IEEE/CAA Journal of Automatica Sinica, vol. 12, no. 10, pp. 2165–2167, 2025

    2025

  25. [25]

    Ncsac: Effective neural community search via attribute-augmented conductance,

    L. Lin, Q. Li, M. Qiao, Z. Wang, J. Zhao, R.-H. Li, X. Luo, and T. Jia, “Ncsac: Effective neural community search via attribute-augmented conductance,”IEEE Transactions on Knowledge and Data Engineering, vol. 38, no. 2, pp. 1221–1235, 2026

    2026

  26. [26]

    Link-based attributed graph clustering via approximate generative bayesian learning,

    Y . Yang, L. Hu, G. Li, D. Li, P. Hu, and X. Luo, “Link-based attributed graph clustering via approximate generative bayesian learning,”IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 55, no. 8, pp. 5730–5743, 2025

    2025

  27. [27]

    Graph-based prediction of mirna-drug associations with multisource information and metapath enhancement matrices,

    M.-Y . Wu, P. Hu, Z.-H. You, J. Zhang, L. Hu, and X. Luo, “Graph-based prediction of mirna-drug associations with multisource information and metapath enhancement matrices,”IEEE Journal of Biomedical and Health Informatics, 2025

    2025

  28. [28]

    A fast nonnegative autoencoder-based approach to latent feature analysis on high-dimensional and incomplete data,

    F. Bi, T. He, and X. Luo, “A fast nonnegative autoencoder-based approach to latent feature analysis on high-dimensional and incomplete data,”IEEE Transactions on Services Computing, vol. 17, no. 3, pp. 733–746, 2024

    2024

  29. [29]

    A novel approach to temporal qos estimation via extended kalman filter-incorporated latent feature analysis,

    Y . Yuan, S. Wang, H. Zhou, L. Wang, and X. Luo, “A novel approach to temporal qos estimation via extended kalman filter-incorporated latent feature analysis,”IEEE Transactions on Services Computing, 2026

    2026

  30. [30]

    A robust approach to electricity theft detection via tensor representation-driven contrastive distillation,

    W. Qin, Y . Ding, and X. Luo, “A robust approach to electricity theft detection via tensor representation-driven contrastive distillation,”IEEE Transactions on Industrial Informatics, 2026

    2026

  31. [31]

    Temporal web service qos prediction via kalman filter-incorporated dynamic latent factor analysis,

    Y . Yuan, M. Shang, and X. Luo, “Temporal web service qos prediction via kalman filter-incorporated dynamic latent factor analysis,” inEuropean Conference on Artificial Intelligence, 2020, pp. 561–568

    2020

  32. [32]

    A sampling-neighborhood-regularized latent factorization of tensor for dynamic qos estimation,

    X. Xu, M. Lin, Z. Xu, and X. Luo, “A sampling-neighborhood-regularized latent factorization of tensor for dynamic qos estimation,”IEEE Transactions on Network and Service Management, vol. 23, pp. 1707–1722, 2026

    2026

  33. [33]

    A novel tensor causal convolution network model for highly-accurate representation to spatio-temporal data,

    X. Liao, H. Wu, and X. Luo, “A novel tensor causal convolution network model for highly-accurate representation to spatio-temporal data,”IEEE Transactions on Automation Science and Engineering, vol. 22, pp. 19 525–19 537, 2025

    2025

  34. [34]

    An adaptively bias-extended non-negative latent factorization of tensors model for accurately representing the dynamic qos data,

    X. Xu, M. Lin, X. Luo, and Z. Xu, “An adaptively bias-extended non-negative latent factorization of tensors model for accurately representing the dynamic qos data,”IEEE Transactions on Services Computing, vol. 18, no. 2, pp. 603–617, 2025

    2025

  35. [35]

    Multi-aspect self-attending neural tucker factorization for spatiotemporal representation learning,

    Y . Hou, P. Tang, and X. Luo, “Multi-aspect self-attending neural tucker factorization for spatiotemporal representation learning,”IEEE/CAA Journal of Automatica Sinica, 2025

    2025

  36. [36]

    Latent factorization of tensors incorporated battery cycle life prediction,

    M. Chen, L. Tao, J. Lou, and X. Luo, “Latent factorization of tensors incorporated battery cycle life prediction,”IEEE/CAA Journal of Automatica Sinica, vol. 12, no. 3, pp. 633–635, 2025

    2025

  37. [37]

    Asynchronous parallel fuzzy stochastic gradient descent for high-dimensional incomplete data,

    W. Qin and X. Luo, “Asynchronous parallel fuzzy stochastic gradient descent for high-dimensional incomplete data,”IEEE Transactions on Fuzzy Systems, vol. 32, no. 2, pp. 445–459, 2024

    2024

  38. [38]

    Discovering spatio-temporal-individual coupled features from nonstandard tensors-a novel dynamic graph mixer approach,

    F. Bi, T. He, Y .-S. Ong, and X. Luo, “Discovering spatio-temporal-individual coupled features from nonstandard tensors-a novel dynamic graph mixer approach,”IEEE Transactions on Neural Networks and Learning Systems, vol. 36, no. 11, pp. 19 834–19 848, 2025

    2025

  39. [39]

    Sgd-dyg: Self-reliant global dependency apprehending on dynamic graphs,

    M. Han, L. Wang, Y . Yuan, and X. Luo, “Sgd-dyg: Self-reliant global dependency apprehending on dynamic graphs,” inACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2025, pp. 802–813

    2025

  40. [40]

    Knowledge-driven multiple instance learning with hierarchical cluster-incorporated aware filtering for larynx pathological grading,

    C. Li, P. Huang, J. Qin, and X. Luo, “Knowledge-driven multiple instance learning with hierarchical cluster-incorporated aware filtering for larynx pathological grading,”IEEE Journal of Biomedical and Health Informatics, 2025

    2025

  41. [41]

    Fmvpci: A multi-view fusion neural network for identifying protein complex via fuzzy clustering,

    Y . Yang, L. Hu, G. Li, D. Li, P. Hu, and X. Luo, “Fmvpci: A multi-view fusion neural network for identifying protein complex via fuzzy clustering,” IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 55, no. 9, pp. 6189–6202, 2025

    2025

  42. [42]

    Fuzzy mixture-of-experts aggregation for organoid identification with multi-scale state space features,

    X. Deng, P. Hu, T. Herget, F. Tan, X. Zhu, J. Zhang, Y .-a. Huang, L. Hu, Z. You, and X. Luo, “Fuzzy mixture-of-experts aggregation for organoid identification with multi-scale state space features,”IEEE Transactions on Fuzzy Systems, vol. 34, no. 1, pp. 324–335, 2026

    2026

  43. [43]

    Analysis of students’ positive emotion and smile intensity using sequence-relative key-frame labeling and deep-asymmetric convolutional neural network,

    Z. Luo, X. Jin, Y . Luo, Q. Zhou, and X. Luo, “Analysis of students’ positive emotion and smile intensity using sequence-relative key-frame labeling and deep-asymmetric convolutional neural network,”IEEE/CAA Journal of Automatica Sinica, vol. 12, no. 4, pp. 806–820, 2025

    2025

  44. [44]

    Multi metric autoencoder for representing high-dimensional and incomplete data,

    D. Wu, C. Liang, Y . He, Y . Qiao, and X. Luo, “Multi metric autoencoder for representing high-dimensional and incomplete data,”IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2025

    2025

  45. [45]

    Symmetry and graph bi-regularized non-negative matrix factorization for precise community detection,

    Z. Liu, X. Luo, and M. Zhou, “Symmetry and graph bi-regularized non-negative matrix factorization for precise community detection,”IEEE Transactions on Automation Science and Engineering, vol. 21, no. 2, pp. 1406–1420, 2024

    2024

  46. [46]

    Learning accurate representation to nonstandard tensors via a mode-aware tucker network,

    H. Wu, Q. Wang, X. Luo, and Z. Wang, “Learning accurate representation to nonstandard tensors via a mode-aware tucker network,”IEEE Transactions on Knowledge and Data Engineering, vol. 37, no. 12, pp. 7272–7285, 2025

    2025

  47. [47]

    Dynamic stochastic reorientation particle swarm optimization for adaptive latent factor analysis in high- dimensional sparse matrices,

    C. Lyu, Z. Ma, X. Luo, and Y . Shi, “Dynamic stochastic reorientation particle swarm optimization for adaptive latent factor analysis in high- dimensional sparse matrices,”IEEE Transactions on Knowledge and Data Engineering, vol. 38, no. 1, pp. 222–234, 2026

    2026

  48. [48]

    Adaptive pid-incorporated nonnegative latent factor analysis,

    J. Li, Y . Yuan, T. He, and X. Luo, “Adaptive pid-incorporated nonnegative latent factor analysis,”IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2026

    2026

  49. [49]

    Non-gradient hash factor learning for high-dimensional and incomplete data representation learning,

    D. Wu, S. Li, Y . He, X. Luo, and X. Gao, “Non-gradient hash factor learning for high-dimensional and incomplete data representation learning,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2026

    2026

  50. [50]

    A kalman-filter-incorporated latent factor analysis model for temporally dynamic sparse data,

    Y . Yuan, X. Luo, M. Shang, and Z. Wang, “A kalman-filter-incorporated latent factor analysis model for temporally dynamic sparse data,”IEEE Transactions on Cybernetics, vol. 53, no. 9, pp. 5788–5801, 2023

    2023

  51. [51]

    Neural non-negative latent factorization of tensors model with acceleration and unconstraint,

    W. Li, M. Lin, X. Xu, L. Lin, Z. Xu, and X. Luo, “Neural non-negative latent factorization of tensors model with acceleration and unconstraint,” IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 56, no. 1, pp. 164–178, 2026

    2026

  52. [52]

    An outlier-resilient autoencoder for representing high-dimensional and incomplete data,

    D. Wu, Y . Hu, K. Liu, J. Li, X. Wang, S. Deng, N. Zheng, and X. Luo, “An outlier-resilient autoencoder for representing high-dimensional and incomplete data,”IEEE Transactions on Emerging Topics in Computational Intelligence, vol. 9, no. 2, pp. 1379–1391, 2025

    2025

  53. [53]

    A prediction-sampling-based multilayer-structured latent factor model for accurate representation to high- dimensional and sparse data,

    D. Wu, X. Luo, Y . He, and M. Zhou, “A prediction-sampling-based multilayer-structured latent factor model for accurate representation to high- dimensional and sparse data,”IEEE Transactions on Neural Networks and Learning Systems, vol. 35, no. 3, pp. 3845–3858, 2024

    2024

  54. [54]

    Poskhg: A position-aware knowledge hypergraph model for link prediction,

    Z. Chen, X. Wang, C. Wang, and Z. Li, “Poskhg: A position-aware knowledge hypergraph model for link prediction,”Data Science and Engineering, vol. 8, no. 2, pp. 135–145, 2023

    2023

  55. [55]

    Few-shot relation prediction of knowledge graph via convolutional neural network with self-attention,

    S. Zhong, J. Wang, K. Yue, L. Duan, Z. Sun, and Y . Fang, “Few-shot relation prediction of knowledge graph via convolutional neural network with self-attention,”Data Science and Engineering, vol. 8, no. 4, pp. 385–395, 2023

    2023

  56. [56]

    Label-aware chinese event detection with heterogeneous graph attention network,

    S.-Y . Cui, B.-W. Yu, X. Cong, T.-W. Liu, Q.-F. Tan, and J.-Q. Shi, “Label-aware chinese event detection with heterogeneous graph attention network,” Journal of Computer Science and Technology, vol. 39, no. 1, pp. 227–242, 2024

    2024

  57. [57]

    Meta-learning based few-shot link prediction for emerging knowledge graph,

    Y .-F. Zhang, W. Chen, P.-P. Zhao, J.-J. Xu, J.-H. Fang, and L. Zhao, “Meta-learning based few-shot link prediction for emerging knowledge graph,” Journal of Computer Science and Technology, vol. 39, no. 5, pp. 1058–1077, 2024

    2024

  58. [58]

    A novel graph oversampling framework for node classification in class-imbalanced graphs,

    R. Xia, C. Zhang, Y . Zhang, X. Liu, and B. Yang, “A novel graph oversampling framework for node classification in class-imbalanced graphs,” Science China Information Sciences, vol. 67, no. 6, pp. 1–16, 2024

    2024

  59. [59]

    Multiple types of disease-associated rnas identification for disease prognosis and therapy using heterogeneous graph learning,

    W. Zhang, H. Wei, W. Zhang, H. Wu, and B. Liu, “Multiple types of disease-associated rnas identification for disease prognosis and therapy using heterogeneous graph learning,”Science China Information Sciences, vol. 67, no. 8, p. 189103, 2024

    2024

  60. [60]

    A convolution bias-incorporated nonnegative latent factorization of tensors model for accurate representation learning to dynamic directed graphs,

    Q. Wang, H. Wu, and X. Luo, “A convolution bias-incorporated nonnegative latent factorization of tensors model for accurate representation learning to dynamic directed graphs,”IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 55, no. 12, pp. 8902–8914, 2025

    2025

  61. [61]

    Structure-preserved self-attention for fusion image information in multiple color spaces,

    Z. He, M. Lin, X. Luo, and Z. Xu, “Structure-preserved self-attention for fusion image information in multiple color spaces,”IEEE Transactions on Neural Networks and Learning Systems, vol. 36, no. 7, pp. 13 021–13 035, 2025

    2025

  62. [62]

    A generalized nesterov’s accelerated gradient-incorporated non-negative latent-factorization-of-tensors model for efficient representation to dynamic qos data,

    M. Chen, Y . Qiao, R. Wang, and X. Luo, “A generalized nesterov’s accelerated gradient-incorporated non-negative latent-factorization-of-tensors model for efficient representation to dynamic qos data,”IEEE Transactions on Emerging Topics in Computational Intelligence, vol. 8, no. 3, pp. 2386–2400, 2024

    2024

  63. [63]

    Graph condensation for graph neural networks,

    W. Jin, L. Zhao, S. Zhang, Y . Liu, J. Tang, and N. Shah, “Graph condensation for graph neural networks,” inThe Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29, 2022, 2022

    2022

  64. [64]

    Learning error refinement in stochastic gradient descent-based latent factor analysis via diversified pid controllers,

    J. Li, Y . Yuan, and X. Luo, “Learning error refinement in stochastic gradient descent-based latent factor analysis via diversified pid controllers,” IEEE Transactions on Emerging Topics in Computational Intelligence, vol. 9, no. 5, pp. 3582–3597, 2025

    2025

  65. [65]

    A fuzzy pid-incorporated stochastic gradient descent algorithm for fast and accurate latent factor analysis,

    Y . Yuan, J. Li, and X. Luo, “A fuzzy pid-incorporated stochastic gradient descent algorithm for fast and accurate latent factor analysis,”IEEE Transactions on Fuzzy Systems, vol. 32, no. 7, pp. 4049–4061, 2024

    2024

  66. [66]

    Adaptive divergence-based non-negative latent factor analysis of high-dimensional and incomplete matrices from industrial applications,

    Y . Yuan, X. Luo, and M. Zhou, “Adaptive divergence-based non-negative latent factor analysis of high-dimensional and incomplete matrices from industrial applications,”IEEE Transactions on Emerging Topics in Computational Intelligence, vol. 8, no. 2, pp. 1209–1222, 2024

    2024

  67. [67]

    Genetic algorithm-based two-step optimization for precise latent factor analysis,

    C. Lyu, J. Cheng, X. Luo, and Y . Shi, “Genetic algorithm-based two-step optimization for precise latent factor analysis,”IEEE Transactions on Neural Networks and Learning Systems, 2025

    2025

  68. [68]

    A calibrator fuzzy ensemble for highly-accurate robot arm calibration,

    X. Luo, Z. Li, W. Yue, and S. Li, “A calibrator fuzzy ensemble for highly-accurate robot arm calibration,”IEEE Transactions on Neural Networks and Learning Systems, vol. 36, no. 2, pp. 2169–2181, 2025

    2025

  69. [69]

    Parallel adaptive stochastic gradient descent algorithms for latent factor analysis of high-dimensional and incomplete industrial data,

    W. Qin, X. Luo, S. Li, and M. Zhou, “Parallel adaptive stochastic gradient descent algorithms for latent factor analysis of high-dimensional and incomplete industrial data,”IEEE Transactions on Automation Science and Engineering, vol. 21, no. 3, pp. 2716–2729, 2024

    2024

  70. [70]

    An adaptive divergence-based non-negative latent factor model,

    Y . Yuan, R. Wang, G. Yuan, and X. Luo, “An adaptive divergence-based non-negative latent factor model,”IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 53, no. 10, pp. 6475–6487, 2023

    2023

  71. [71]

    A multilayered-and-randomized latent factor model for high-dimensional and sparse matrices,

    Y . Yuan, Q. He, X. Luo, and M. Shang, “A multilayered-and-randomized latent factor model for high-dimensional and sparse matrices,”IEEE Transactions on Big Data, vol. 8, no. 3, pp. 784–794, 2022

    2022

  72. [72]

    Attention-mechanism-based neural latent-factorization-of-tensors mode,

    X. Xu, M. Lin, Z. Xu, and X. Luo, “Attention-mechanism-based neural latent-factorization-of-tensors mode,”ACM Transactions on Knowledge Discovery from Data, vol. 19, no. 4, pp. 1–27, 2025

    2025

  73. [73]

    A robust coevolutionary neural-based optimization algorithm for constrained nonconvex optimization,

    L. Wei, L. Jin, and X. Luo, “A robust coevolutionary neural-based optimization algorithm for constrained nonconvex optimization,”IEEE Transactions on Neural Networks and Learning Systems, vol. 35, no. 6, pp. 7778–7791, 2024

    2024

  74. [74]

    Heterogeneous graph condensation,

    J. Gao, J. Wu, and J. Ding, “Heterogeneous graph condensation,”IEEE Trans. Knowl. Data Eng., vol. 36, no. 7, pp. 3126–3138, 2024

    2024

  75. [75]

    Are we really making much progress?: Revisiting, benchmarking and refining heterogeneous graph neural networks,

    Q. Lv, M. Ding, Q. Liu, Y . Chen, W. Feng, S. He, C. Zhou, J. Jiang, Y . Dong, and J. Tang, “Are we really making much progress?: Revisiting, benchmarking and refining heterogeneous graph neural networks,” inKDD ’21: The 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2021, pp. 1150–1160

    2021

  76. [76]

    A comprehensive survey on graph reduction: Sparsification, coarsening, and condensation,

    M. Hashemi, S. Gong, J. Ni, W. Fan, B. A. Prakash, and W. Jin, “A comprehensive survey on graph reduction: Sparsification, coarsening, and condensation,”CoRR, vol. abs/2402.03358, 2024

    Pith/arXiv arXiv 2024

  77. [77]

    A proportional integral controller-enhanced non-negative latent factor analysis model,

    Y . Yuan, S. Lu, and X. Luo, “A proportional integral controller-enhanced non-negative latent factor analysis model,”IEEE/CAA Journal of Automatica Sinica, vol. 12, no. 6, pp. 1246–1259, 2025

    2025

  78. [78]

    A scalable multi-channel sentiment analysis model with enhanced semantic understanding and redundancy reduction,

    J. Liu, X. Li, M. Lin, and X. Luo, “A scalable multi-channel sentiment analysis model with enhanced semantic understanding and redundancy reduction,”IEEE Transactions on Computational Social Systems, 2025

    2025

  79. [79]

    Iterative role negotiation via the bi-level gra++ with decision tolerance,

    Q. Jiang, D. Liu, H. Zhu, S. Wu, N. Wu, X. Luo, and Y . Qiao, “Iterative role negotiation via the bi-level gra++ with decision tolerance,”IEEE Transactions on Computational Social Systems, vol. 11, no. 6, pp. 7484–7499, 2024

    2024

  80. [80]

    Spectral sparsification of graphs,

    D. A. Spielman and S. Teng, “Spectral sparsification of graphs,”SIAM J. Comput., vol. 40, no. 4, pp. 981–1025, 2011

    2011

Showing first 80 references.