pith. sign in

arxiv: 2603.26178 · v2 · submitted 2026-03-27 · 💻 cs.LG

Geometric Evolution Graph Convolutional Networks: Enhancing Graph Representation Learning via Ricci Flow

Jicheng Ma , Yunyan Yang , Juan Zhao , Liang Zhao This is my paper

Pith reviewed 2026-05-14 23:55 UTC · model grok-4.3

classification 💻 cs.LG
keywords graph neural networksricci flowgeometric evolutiongraph classificationlstmrepresentation learningdiscrete curvaturegnn
0
0 comments X

The pith

GEGCN enhances graph representation learning by modeling geometric evolution with discrete Ricci flow and LSTM.

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

The paper introduces GEGCN to improve graph neural networks by explicitly incorporating the geometric evolution of graph structures. It applies discrete Ricci flow to create a sequence of evolving graphs, uses an LSTM to learn dynamic representations from this sequence, and combines these with a standard graph convolutional network. Experiments show strong results on node classification for homophilic and heterophilic graphs as well as larger datasets. This matters because standard GCNs treat graphs as static, missing potential benefits from curvature-driven changes in structure that could better capture underlying data geometry.

Core claim

By generating a dynamic structural sequence through discrete Ricci flow and processing it with an LSTM network to obtain learned dynamic representations, which are then infused into a graph convolutional network, GEGCN achieves improved performance on graph classification tasks across a range of benchmark datasets including homophilic, heterophilic, filtered, and large-scale graphs.

What carries the argument

The Geometric Evolution Graph Convolutional Network (GEGCN) that combines discrete Ricci flow for generating evolving graph structures, an LSTM to capture dynamic features from the sequence, and infusion of those features into a GCN.

If this is right

  • GEGCN achieves strong results on both homophilic and heterophilic graphs.
  • The approach maintains performance on filtered graphs and scales to large graphs.
  • Geometric evolution supplies useful signals for node classification beyond static structure.
  • Infusing LSTM-captured flow dynamics into GCN layers produces the reported gains.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same Ricci flow sequence could be tested as a preprocessing step for other GNN tasks like link prediction.
  • Applying the method to graphs known to have strong curvature properties would provide a targeted validation.
  • Replacing discrete Ricci flow with other discrete curvature definitions might produce comparable or stronger results.
  • The framework could extend naturally to time-varying graphs where structural changes occur organically.

Load-bearing premise

That the dynamic structural sequence generated by discrete Ricci flow, when processed by LSTM, supplies representations that meaningfully improve upon a standard graph convolutional network on the tested tasks.

What would settle it

A direct comparison experiment on the same benchmark datasets where GEGCN shows no accuracy gain or performs worse than a plain GCN baseline would falsify the central claim of enhancement.

Figures

Figures reproduced from arXiv: 2603.26178 by Jicheng Ma, Juan Zhao, Liang Zhao, Yunyan Yang.

Figure 1
Figure 1. Figure 1: Overview of the GEGCN framework. It comprises three phases: (i) Geometric Evolution Generator via Discrete Ricci Flow; (ii) Structural Dynamics Encoder for capturing temporal dynamics and constructing the edge importance matrix via LSTM; and (iii) Feature Fusion for integrating geometric insights into curvature-aware GCN layers. Conceptually, while the classical Ricci flow on a smooth manifold defines a co… view at source ↗
Figure 2
Figure 2. Figure 2: Ablation Study. The bar chart compares the mean test accuracy (%) of the baseline GCN, five ablation variants that replace the LSTM module for modeling Ricci curvature evolution with First, Last, Mean Pooling, Max Pooling, or MLP-based aggregation, and the full GEGCN model. The consistent performance gains of GEGCN across all datasets demonstrate the effectiveness of LSTM-based modeling of the Ricci curvat… view at source ↗
Figure 3
Figure 3. Figure 3: Oversmoothing analysis on Cora. We report the mean test accuracy across var￾ious numbers of layers L ∈ {2, 4, 8, 16, 32, 64}. All models eventually collapse to the chance level (30.82%), but GEGCN maintains signifi￾cantly higher accuracy at L = 4 compared to the vanilla GCN. Analysis of Oversmoothing. A well-known drawback of GCNs is that capturing long-range dependencies by adding layers comes at the cost… view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of Ollivier-Ricci curvature in continuous manifolds and discrete graphs. (Top row) Continuous manifolds: (Left) Hyperbolic paraboloid with negative curvature (κ < 0); (Middle) Plane with zero curvature (κ = 0); (Right) Elliptic paraboloid with positive curvature (κ > 0). (Bottom row) Discrete graphs: (Left) Two dense clusters connected by a single edge with negative curvature; (Middle) Regular 3… view at source ↗
read the original abstract

We introduce the Geometric Evolution Graph Convolutional Network (GEGCN), a novel framework that enhances graph representation learning through explicit modeling of geometric evolution on graph structures. Specifically, GEGCN leverages a Long Short-Term Memory (LSTM) network to capture the dynamic structural sequence generated by discrete Ricci flow, and infuses the learned dynamic representations into a graph convolutional network. Extensive experiments demonstrate that GEGCN achieves excellent performance on classification tasks across various benchmark datasets, including homophilic/heterophilic graphs, filtered graphs, and large-scale graphs.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces the Geometric Evolution Graph Convolutional Network (GEGCN), which generates a sequence of evolving graph structures via discrete Ricci flow, encodes the dynamic sequence with an LSTM to produce time-aware representations, and injects these into a graph convolutional network for node classification. The central claim is that this geometric-evolution approach yields excellent performance on classification tasks across homophilic/heterophilic, filtered, and large-scale benchmark graphs.

Significance. If the ablation isolating the Ricci-flow sequence is supplied and the gains are shown to exceed those of a static-graph LSTM-GCN baseline, the work would offer a concrete mechanism for injecting discrete curvature dynamics into GNNs, potentially improving robustness on graphs whose structure evolves or exhibits heterogeneous curvature. No machine-checked proofs or parameter-free derivations are described.

major comments (2)
  1. [Experiments] Experiments section (and any associated tables/figures): the manuscript compares GEGCN against published baselines but does not include the minimal control that keeps the LSTM and GCN components fixed while replacing the discrete-Ricci-flow sequence with either (a) the static adjacency matrix repeated across time steps or (b) a random/non-curvature evolution. Without this ablation the central claim that the geometric evolution itself supplies the performance improvement cannot be evaluated.
  2. [Abstract] Abstract and §1: the claim of 'excellent performance' is stated without any numerical results, metrics, dataset sizes, or baseline deltas, rendering the abstract non-informative for assessing the magnitude or reliability of the reported gains.
minor comments (2)
  1. [Method] Notation for the discrete Ricci-flow update rule and the precise manner in which LSTM hidden states are fused into the GCN layers should be made explicit (e.g., an equation defining the combined representation).
  2. [Figures] Figure captions and axis labels for any evolution-sequence visualizations should state the number of Ricci-flow steps and the curvature threshold used.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address the major comments point by point below and will revise the manuscript to incorporate the suggested improvements.

read point-by-point responses

  1. Referee: [Experiments] Experiments section (and any associated tables/figures): the manuscript compares GEGCN against published baselines but does not include the minimal control that keeps the LSTM and GCN components fixed while replacing the discrete-Ricci-flow sequence with either (a) the static adjacency matrix repeated across time steps or (b) a random/non-curvature evolution. Without this ablation the central claim that the geometric evolution itself supplies the performance improvement cannot be evaluated.

    Authors: We agree that this ablation is required to rigorously isolate the contribution of the discrete Ricci flow. In the revised version we will add two controls that keep the LSTM and GCN components unchanged: (a) the static adjacency matrix repeated across all time steps, and (b) a random non-curvature evolution sequence. The new results will be reported in the experiments section and tables, allowing direct quantification of the performance lift attributable to the geometric evolution. revision: yes

  2. Referee: [Abstract] Abstract and §1: the claim of 'excellent performance' is stated without any numerical results, metrics, dataset sizes, or baseline deltas, rendering the abstract non-informative for assessing the magnitude or reliability of the reported gains.

    Authors: We accept the criticism. The abstract will be rewritten to include concrete numerical results (accuracy and F1 scores on representative datasets), dataset sizes, and explicit baseline deltas so that readers can immediately evaluate the magnitude and reliability of the reported improvements. revision: yes

Circularity Check

0 steps flagged

No significant circularity in modeling framework or claims

full rationale

The paper presents GEGCN as an empirical architecture that generates a sequence of graphs via discrete Ricci flow, encodes the sequence with LSTM, and combines the result with a GCN for node/graph classification. No mathematical derivation chain, first-principles prediction, or fitted parameter is described that reduces by construction to its own inputs. Performance claims rest on benchmark experiments rather than any self-referential equivalence. No load-bearing self-citation, ansatz smuggling, or renaming of known results is evident in the abstract or described method. The contribution is a proposed combination of existing components (Ricci flow, LSTM, GCN) whose value is tested externally on datasets; this structure is self-contained and does not exhibit the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The abstract contains no explicit free parameters, axioms, or invented entities beyond the high-level description of the GEGCN architecture itself.

pith-pipeline@v0.9.0 · 5381 in / 1203 out tokens · 38556 ms · 2026-05-14T23:55:03.449423+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

58 extracted references · 58 canonical work pages

  1. [1]

    Graph Neu- ral Networks: A Review of Methods and Applications

    Zhou J, Cui G, Hu S, et al. Graph Neu- ral Networks: A Review of Methods and Applications. AI Open 2020;1:57–81

    work page 2020

  2. [2]

    Semi-supervised classification with graph convolutional net- works

    Kipf TN and Welling M. Semi-supervised classification with graph convolutional net- works. In:International Conference on Learning Representations. 2017

    work page 2017

  3. [3]

    Graph attention networks

    Veliˇ ckovi´ c P, Cucurull G, Casanova A, Romero A, Lio P, and Bengio Y. Graph attention networks. In:International Con- ference on Learning Representations. 2018

    work page 2018

  4. [4]

    In- ductive representation learning on large graphs

    Hamilton WL, Ying R, and Leskovec J. In- ductive representation learning on large graphs. In:Proceedings of the 31st In- ternational Conference on Neural Infor- mation Processing Systems. Long Beach, California, USA: Curran Associates Inc., 2017:1025–35

    work page 2017

  5. [5]

    Ricci curvature of Markov chains on metric spaces

    Ollivier Y. Ricci curvature of Markov chains on metric spaces. Journal of Func- tional Analysis 2009;256:810–64

    work page 2009

  6. [6]

    Bochner’s method for cell complexes and combinatorial Ricci curva- ture

    Forman R. Bochner’s method for cell complexes and combinatorial Ricci curva- ture. Discrete & Computational Geometry 2003;29:323–74

    work page 2003

  7. [7]

    Ricci curvature of graphs

    Lin Y, Lu L, and Yau ST. Ricci curvature of graphs. Tohoku Mathematical Journal 2011;63:605–27

    work page 2011

  8. [8]

    Under- standing over-squashing and bottlenecks on graphs

    Topping J, Di Giovanni F, Chamberlain BP, Dong X, and Bronstein MM. Under- standing over-squashing and bottlenecks on graphs. In:International Conference on Learning Representations. 2022

    work page 2022

  9. [9]

    Com- munity Detection on Networks with Ricci Flow

    Ni CC, Lin YY, Luo F, and Gao J. Com- munity Detection on Networks with Ricci Flow. Scientific Reports 2019;9:9984

    work page 2019

  10. [10]

    Normalized Dis- crete Ricci Flow used in Community De- tection

    Lai X, Bai S, and Lin Y. Normalized Dis- crete Ricci Flow used in Community De- tection. Physica A: Statistical Mechanics and its Applications 2022;597:127251

    work page 2022

  11. [11]

    Graph Curvature for Network Fragility: An Application to Financial Markets

    Sandhu RS, Georgiou TT, Reznik E, et al. Graph Curvature for Network Fragility: An Application to Financial Markets. Science Advances 2016;2:e1501495

    work page 2016

  12. [12]

    Curvature Graph Network

    Ye Z, Liu KS, Ma T, Gao J, and Chen C. Curvature Graph Network. In:Interna- tional Conference on Learning Representa- tions. 2020

    work page 2020

  13. [13]

    CurvDrop: A Ricci Curvature Based Approach to Pre- vent Graph Neural Networks from Over- Smoothing and Over-Squashing

    Liu Y, Zhou C, Pan S, et al. CurvDrop: A Ricci Curvature Based Approach to Pre- vent Graph Neural Networks from Over- Smoothing and Over-Squashing. In:Pro- ceedings of the ACM Web Conference 2023. WWW ’23. Austin, TX, USA, 2023:221– 30

    work page 2023

  14. [14]

    Ollivier-Ricci curvature and the spectrum of the normal- ized graph Laplace operator

    Bauer F, Jost J, and Liu S. Ollivier-Ricci curvature and the spectrum of the normal- ized graph Laplace operator. Mathematical Research Letters 2012;19:1185–205

    work page 2012

  15. [15]

    Ollivier Ricci-flow on weighted graphs

    Bai S, Lin Y, Lu L, Wang Z, and Yau ST. Ollivier Ricci-flow on weighted graphs. American Journal of Mathematics 2024;146:1723–47

    work page 2024

  16. [16]

    A modified Ricci flow on arbitrary weighted graph

    Ma J and Yang Y. A modified Ricci flow on arbitrary weighted graph. The Journal of Geometric Analysis 2025;35:332

    work page 2025

  17. [17]

    Net- work Alignment by Discrete Ollivier-Ricci Flow

    Ni CC, Lin YY, Gao J, and Gu X. Net- work Alignment by Discrete Ollivier-Ricci Flow. In:Graph Drawing and Network Vi- 14 sualization. Cham: Springer International Publishing, 2018:447–62

    work page 2018

  18. [18]

    Community detection of undirected hyper- graphs by Ricci flow

    Tian Y, Ma J, Yang Y, and Zhao L. Community detection of undirected hyper- graphs by Ricci flow. Physical Review E 4 2025;112:044311

    work page 2025

  19. [19]

    Long Short-Term Memory

    Hochreiter S and Schmidhuber J. Long Short-Term Memory. Neural Computation 1997;9:1735–80

    work page 1997

  20. [20]

    Curvature Graph Neural Network

    Li H, Cao J, Zhu J, Liu Y, Zhu Q, and Wu G. Curvature Graph Neural Network. Information Sciences 2022;592:50–66

    work page 2022

  21. [21]

    κHGCN: Tree-likeness Modeling via Con- tinuous and Discrete Curvature Learning

    Yang M, Zhou M, Pan L, and King I. κHGCN: Tree-likeness Modeling via Con- tinuous and Discrete Curvature Learning. In:Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. KDD ’23. Long Beach, CA, USA: Association for Computing Machin- ery, 2023:2965–77

    work page 2023

  22. [22]

    A Self- supervised Riemannian GNN with Time Varying Curvature for Temporal Graph Learning

    Sun L, Ye J, Peng H, and Yu PS. A Self- supervised Riemannian GNN with Time Varying Curvature for Temporal Graph Learning. In:Proceedings of the 31st ACM International Conference on Information & Knowledge Management. CIKM ’22. At- lanta, GA, USA: Association for Comput- ing Machinery, 2022:1827–36

    work page 2022

  23. [23]

    How curvature enhance the adaptation power of framelet GCNs

    Shi D, Guo Y, Shao Z, and Gao J. How curvature enhance the adaptation power of framelet GCNs. arXiv preprint arXiv:2307.09768 2023

    work page arXiv 2023

  24. [24]

    Revisiting Over-smoothing and Over-squashing Us- ing Ollivier-Ricci Curvature

    Nguyen K, Hieu NM, Nguyen VD, Ho N, Osher S, and Nguyen TM. Revisiting Over-smoothing and Over-squashing Us- ing Ollivier-Ricci Curvature. In:Proceed- ings of the 40th International Conference on Machine Learning. Vol. 202. Proceed- ings of Machine Learning Research. PMLR, 2023:25956–79

    work page 2023

  25. [25]

    On the Trade-off between Over-smoothing and Over-squashing in Deep Graph Neural Networks

    Giraldo JH, Skianis K, Bouwmans T, and Malliaros FD. On the Trade-off between Over-smoothing and Over-squashing in Deep Graph Neural Networks. In:Pro- ceedings of the 32nd ACM International Conference on Information and Knowl- edge Management. CIKM ’23. Birmingham, United Kingdom: Association for Comput- ing Machinery, 2023:566–76

    work page 2023

  26. [26]

    Discrete Cur- vature Graph Information Bottleneck

    Fu X, Wang J, Gao Y, et al. Discrete Cur- vature Graph Information Bottleneck. Pro- ceedings of the AAAI Conference on Arti- ficial Intelligence 2025;39:16666–73

    work page 2025

  27. [27]

    GRAND: Graph Neural Diffusion

    Chamberlain B, Rowbottom J, Gorinova MI, Bronstein M, Webb S, and Rossi E. GRAND: Graph Neural Diffusion. In:In- ternational Conference on Machine Learn- ing (ICML) / PMLR. Vol. 139. 2021:1407– 18

    work page 2021

  28. [28]

    Beltrami flow and neural diffusion on graphs

    Chamberlain BP, Rowbottom J, Eynard D, Di Giovanni F, Dong X, and Bronstein MM. Beltrami flow and neural diffusion on graphs. In:Proceedings of the 35th Interna- tional Conference on Neural Information Processing Systems. Vol. 34. Red Hook, NY, USA: Curran Associates Inc., 2021:1594– 609

    work page 2021

  29. [29]

    Graph Neural Ricci Flow: Evolv- ing Feature from a Curvature Perspective

    Chen J, Deng B, WANG Z, Chen C, and Zheng Z. Graph Neural Ricci Flow: Evolv- ing Feature from a Curvature Perspective. In:International Conference on Learning Representations. 2025

    work page 2025

  30. [30]

    Untersuchungen ¨ uber allge- meine Metrik

    Menger K. Untersuchungen ¨ uber allge- meine Metrik. Mathematische Annalen 1930;103:466–501

    work page 1930

  31. [31]

    Discrete geometry: curvature in abstract metric spaces

    Haantjes J. Discrete geometry: curvature in abstract metric spaces. Proc. Kon. Ned. Akad. v. Wetenseh. 1947;50:302–14

    work page 1947

  32. [32]

    Hypercontractivit´ e de semigroupes de diffusion

    Bakry D and ´Emery M. Hypercontractivit´ e de semigroupes de diffusion. C. R. Acad. Sci. Paris S´ er. I Math. 1984;299:775–8. 15

    work page 1984

  33. [33]

    Collective Classification in Network Data

    Sen P, Namata G, Bilgic M, Getoor L, Gallagher B, and Eliassi-Rad T. Collective Classification in Network Data. AI Maga- zine 2008;29:93–106

    work page 2008

  34. [34]

    Image-Based Recommendations on Styles and Substitutes

    McAuley J, Targett C, Shi Q, and Hengel A van den. Image-Based Recommendations on Styles and Substitutes. In:Proceedings of the 38th International ACM SIGIR Con- ference on Research and Development in Information Retrieval. SIGIR ’15. Santiago, Chile, 2015:43–52

    work page 2015

  35. [35]

    Learning to extract symbolic knowledge from the World Wide Web

    Craven M, DiPasquo D, Freitag D, et al. Learning to extract symbolic knowledge from the World Wide Web. In:Proceedings of the Fifteenth National/Tenth Conference on Artificial Intelligence/Innovative Ap- plications of Artificial Intelligence. AAAI ’98/IAAI ’98. Madison, Wisconsin, USA, 1998:509–16

    work page 1998

  36. [36]

    Multi-scale attributed node embed- ding

    Rozemberczki B, Allen C, and Sarkar R. Multi-scale attributed node embed- ding. Journal of Complex Networks 2021;9:cnab014

    work page 2021

  37. [37]

    Social influence analysis in large-scale networks

    Tang J, Sun J, Wang C, and Yang Z. Social influence analysis in large-scale networks. In:Proceedings of the 15th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. KDD ’09. Paris, France: Association for Computing Machinery, 2009:807–16

    work page 2009

  38. [38]

    A crit- ical look at the evaluation of GNNs un- der heterophily: Are we really making progress? In:The Eleventh International Conference on Learning Representations

    Platonov O, Kuznedelev D, Diskin M, Babenko A, and Prokhorenkova L. A crit- ical look at the evaluation of GNNs un- der heterophily: Are we really making progress? In:The Eleventh International Conference on Learning Representations. 2023

    work page 2023

  39. [39]

    Open Graph Benchmark: Datasets for Machine Learn- ing on Graphs

    Hu W, Fey M, Zitnik M, et al. Open Graph Benchmark: Datasets for Machine Learn- ing on Graphs. In:Advances in Neural Information Processing Systems. Vol. 33. Curran Associates, Inc., 2020:22118–33

    work page 2020

  40. [40]

    Rep- resentation learning on graphs with jump- ing knowledge networks

    Xu K, Li C, Tian Y, Sonobe T, Kawarabayashi Ki, and Jegelka S. Rep- resentation learning on graphs with jump- ing knowledge networks. In:International Conference on Machine Learning (ICML) / PMLR. 2018:5453–62

    work page 2018

  41. [41]

    Predict then propa- gate: Graph neural networks meet personalized pagerank

    Gasteiger J, Bojchevski A, and G¨ unnemann S. Predict then propa- gate: Graph neural networks meet personalized pagerank. In:International Conference on Learning Representations. 2019

    work page 2019

  42. [42]

    Adaptive universal generalized pagerank graph neural network

    Chien E, Peng J, Li P, and Milenkovic O. Adaptive universal generalized pagerank graph neural network. In:International Conference on Learning Representations. 2021

    work page 2021

  43. [43]

    Geomet- ric deep learning on graphs and manifolds using mixture model cnns

    Monti F, Boscaini D, Masci J, Rodola E, Svoboda J, and Bronstein MM. Geomet- ric deep learning on graphs and manifolds using mixture model cnns. In:Proceedings of the IEEE conference on computer vision and pattern recognition. 2017:5115–24

    work page 2017

  44. [44]

    How Framelets Enhance Graph Neural Net- works

    Zheng X, Zhou B, Gao J, et al. How Framelets Enhance Graph Neural Net- works. In:International Conference on Machine Learning (ICML) / PMLR. 2021:12761–71

    work page 2021

  45. [45]

    Diffusion improves graph learning

    Gasteiger J, Weißenberger S, and G¨ unnemann S. Diffusion improves graph learning. In:Proceedings of the 33rd International Conference on Neural Information Processing Systems. Red Hook, NY, USA: Curran Associates Inc., 2019:13366–78

    work page 2019

  46. [46]

    On the bottleneck of graph neural networks and its practical implications

    Alon U and Yahav E. On the bottleneck of graph neural networks and its practical implications. In:International Conference on Learning Representations. 2021

    work page 2021

  47. [47]

    Deep Residual Learning for Image Recognition

    He K, Zhang X, Ren S, and Sun J. Deep Residual Learning for Image Recognition. In:2016 IEEE Conference on Computer 16 Vision and Pattern Recognition (CVPR). 2016:770–8

    work page 2016

  48. [48]

    Beyond homophily in graph neural networks: current limitations and effective designs

    Zhu J, Yan Y, Zhao L, Heimann M, Akoglu L, and Koutra D. Beyond homophily in graph neural networks: current limitations and effective designs. In:Proceedings of the 34th International Conference on Neu- ral Information Processing Systems. NIPS ’20. Vancouver, BC, Canada: Curran Asso- ciates Inc., 2020

    work page 2020

  49. [49]

    Finding Global Homophily in Graph Neural Net- works When Meeting Heterophily

    Li X, Zhu R, Cheng Y, et al. Finding Global Homophily in Graph Neural Net- works When Meeting Heterophily. In:Pro- ceedings of the 39th International Confer- ence on Machine Learning. Vol. 162. Pro- ceedings of Machine Learning Research. PMLR, 2022:13242–56

    work page 2022

  50. [50]

    Sim- plifying approach to node classification in Graph Neural Networks

    Maurya SK, Liu X, and Murata T. Sim- plifying approach to node classification in Graph Neural Networks. Journal of Com- putational Science 2022;62:101695

    work page 2022

  51. [51]

    Be- yond Low-frequency Information in Graph Convolutional Networks

    Bo D, Wang X, Shi C, and Shen H. Be- yond Low-frequency Information in Graph Convolutional Networks. In:AAAI. AAAI Press, 2021

    work page 2021

  52. [52]

    GBK-GNN: Gated Bi-Kernel Graph Neural Networks for Modeling Both Homophily and Het- erophily

    Du L, Shi X, Fu Q, et al. GBK-GNN: Gated Bi-Kernel Graph Neural Networks for Modeling Both Homophily and Het- erophily. In:Proceedings of the ACM Web Conference 2022. WWW ’22. Virtual Event, Lyon, France: Association for Com- puting Machinery, 2022:1550–8

    work page 2022

  53. [53]

    Node2Vec: Scal- able feature learning for networks

    Grover A and Leskovec J. Node2Vec: Scal- able feature learning for networks. In:Pro- ceedings of the 22nd ACM SIGKDD inter- national conference on Knowledge discov- ery and data mining. 2016:855–64

    work page 2016

  54. [54]

    Masked Label Prediction: Unified Message Passing Model for Semi- Supervised Classification

    Shi Y, Huang Z, Feng S, Zhong H, Wang W, and Sun Y. Masked Label Prediction: Unified Message Passing Model for Semi- Supervised Classification. In:Proceedings of the Thirtieth International Joint Con- ference on Artificial Intelligence, IJCAI-

    work page

  55. [55]

    by Zhou ZH

    Ed. by Zhou ZH. International Joint Conferences on Artificial Intelligence Or- ganization, 2021:1548–54

    work page 2021

  56. [56]

    Adaptive graph diffusion net- works: compact and expressive GNNs with large receptive fields

    Sun C, Zhang M, Hu J, Gu H, Chen J, and Yang M. Adaptive graph diffusion net- works: compact and expressive GNNs with large receptive fields. Artificial Intelligence Review 2025;58:Article number 107

    work page 2025

  57. [57]

    Sinkhorn distances: lightspeed computation of optimal transport

    Cuturi M. Sinkhorn distances: lightspeed computation of optimal transport. In:Pro- ceedings of the 27th International Confer- ence on Neural Information Processing Sys- tems. Vol. 2. Lake Tahoe, Nevada: Curran Associates Inc., 2013:2292–300

    work page 2013

  58. [58]

    Geom-GCN: Geometric Graph Convolutional Networks

    Pei H, Wei B, Chang KWC, Lei Y, and Yang B. Geom-GCN: Geometric Graph Convolutional Networks. In:International Conference on Learning Representations. 2020. 17 A Discrete Ricci Curvature and Ricci Flow A.1 Summary of Notation Notation Description G= (V,E) Graph with node setVand edge setE we Weight of edgee∈ E κe Ricci curvature of edgee∈ E ρe Distance be...

    work page 2020