pith. sign in

arxiv: 2606.06397 · v2 · pith:QQK6CJTLnew · submitted 2026-06-04 · 💻 cs.LG

The Post-GCN Decade Revisited: Curvature-Stratified Evaluation of Relational Learning

Shuo Wang , Xiangyu Wang , Quanxin Wang , Bailin Wu , Bokui Wang , Shunyang Huang , Boyan Deng , Haonan Liu
show 4 more authors
Ruiyi Fang Zhenxiang Xu Boyu Wang Zhao Kang
This is my paper

Pith reviewed 2026-06-28 02:37 UTC · model grok-4.3

classification 💻 cs.LG
keywords relational learninggraph neural networkscurvatureevaluation frameworkgraph convolutional networksmodel generalizationgeometric properties
0
0 comments X

The pith

Relational learning model rankings depend on dataset curvature rather than being universal.

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

Current evaluation practices average performance across datasets assuming uniform structure, which introduces bias. The paper shows that stratifying datasets by their intrinsic curvature reveals stable model rankings within each regime but significant shifts across positive, negative, and near-zero curvature regimes. This indicates that model effectiveness is geometry-dependent. The work evaluates 18 models on 14 datasets using a new curvature-stratified framework and proposes a geometry-aware evaluation protocol for more reliable comparisons.

Core claim

Model rankings are highly stable within each curvature regime but shift significantly across regimes, indicating that performance is fundamentally geometry-dependent rather than universally transferable.

What carries the argument

Curvature-stratified evaluation framework that partitions datasets into positive, negative, and near-zero curvature regimes.

If this is right

  • Model rankings remain consistent when datasets share the same curvature regime.
  • Performance comparisons across different curvature regimes can lead to misleading conclusions about model superiority.
  • Graph foundation models show diminishing returns in certain curvature regimes compared to geometry-aligned graph neural networks.
  • Geometry-aware evaluation provides more interpretable and reliable model assessments than aggregated benchmarks.

Where Pith is reading between the lines

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

  • Future model development could benefit from explicitly incorporating curvature awareness in architectures.
  • Dataset selection for benchmarks should prioritize diversity in geometric properties to avoid biased evaluations.
  • Similar stratification approaches might apply to other latent structural factors beyond curvature.

Load-bearing premise

Curvature serves as the primary latent geometric factor that governs model effectiveness across datasets.

What would settle it

If model rankings do not shift when comparing performance across datasets from different curvature regimes, the claim of geometry-dependent performance would be falsified.

Figures

Figures reproduced from arXiv: 2606.06397 by Bailin Wu, Bokui Wang, Boyan Deng, Boyu Wang, Haonan Liu, Quanxin Wang, Ruiyi Fang, Shunyang Huang, Shuo Wang, Xiangyu Wang, Zhao Kang, Zhenxiang Xu.

Figure 1
Figure 1. Figure 1: Rank-shift heatmap. Thus, both κ¯(G) and γκ(G) are needed to explain performance; a flat average obscures this special￾ist–robustness trade-off.iled curvature profiles, where a small subset of nodes or edges carries disproportionately large geometric signal. To sum up, [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Total training time heatmap on Node Classification (NC) task [PITH_FULL_IMAGE:figures/full_fig_p032_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Mirrored efficiency diagram across models. [PITH_FULL_IMAGE:figures/full_fig_p032_4.png] view at source ↗
read the original abstract

Current evaluation practices in relational learning rely heavily on flat leaderboards that average performance across heterogeneous datasets, implicitly assuming a uniform underlying structure. We show that this assumption introduces systematic bias: it obscures geometry-dependent performance variations and can lead to misleading conclusions about model generalization. In this work, we identify intrinsic geometry as a key latent factor governing model effectiveness. We demonstrate that conventional aggregated metrics mask critical performance trade-offs that only become visible when datasets are stratified by their geometric properties. To address this issue, we introduce a curvature-stratified evaluation framework that partitions datasets into positive, negative, and near-zero curvature regimes. Our benchmark evaluates 18 representative models including Graph Convolutional Networks (GCNs), Graph Foundation Models (GFMs), and tabular learning methods across 14 datasets. We find that model rankings are highly stable within each curvature regime but shift significantly across regimes, indicating that performance is fundamentally geometry-dependent rather than universally transferable. Notably, we identify regimes where GFMs offer diminishing returns compared to geometry-aligned GNNs. Based on these findings, we propose a geometry-aware evaluation protocol that yields more reliable and interpretable comparisons than standard aggregated benchmarks. We release all code, curvature-stratified dataset splits, and evaluation tools to support reproducible and rigorous assessment of future relational learning methods. Code and datasets are provided in our project homepage: https://sirbabbage.github.io/CurvBench_HOME/.

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

3 major / 2 minor

Summary. The manuscript claims that conventional aggregated leaderboards in relational learning introduce systematic bias by averaging over heterogeneous datasets and ignoring intrinsic geometry. It introduces a curvature-stratified evaluation framework that partitions 14 datasets into positive, negative, and near-zero curvature regimes, evaluates 18 models (including GCNs, GFMs, and tabular methods), and reports that model rankings are stable within each regime but shift significantly across regimes. This is taken to indicate that performance is fundamentally geometry-dependent; the work proposes a geometry-aware evaluation protocol and releases code, curvature-stratified splits, and tools.

Significance. If the stratification isolates curvature as the primary factor, the results would support a shift from flat leaderboards toward geometry-aware benchmarks, with practical implications for model selection (e.g., regimes where GFMs yield diminishing returns). The release of code, splits, and evaluation tools is a clear strength that enables reproducibility and extension by others.

major comments (3)
  1. [Abstract and §4] Abstract and §4: the claim that rankings 'shift significantly across regimes' is presented without any statistical tests (e.g., rank-correlation p-values, permutation tests on rank differences, or bootstrap intervals), so it is impossible to determine whether the observed changes exceed sampling variability.
  2. [§3] §3 (Curvature Stratification) and methods: the partitioning into positive/negative/near-zero regimes depends on unspecified thresholds that are free parameters; no sensitivity analysis or justification is provided, so the reported within-regime stability and cross-regime shifts may be artifacts of these cutoffs.
  3. [§4] §4 (Experiments): no controls, matching, or multivariate regression are described for potential confounders (node/edge count, density, average degree, homophily, task type) that may covary with curvature regime; without such checks the attribution of ranking changes specifically to geometry rather than these covariates remains unverified.
minor comments (2)
  1. [§4] A summary table listing each of the 14 datasets, its computed curvature, assigned regime, and basic statistics (size, density) would make the stratification transparent.
  2. [§3] The precise discrete curvature formula employed (e.g., Forman, Ollivier, or other) and any implementation details or verification steps should be stated explicitly in the methods.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback. The comments correctly identify opportunities to strengthen the statistical support for our claims and to verify that curvature, rather than correlated dataset properties, drives the observed ranking shifts. We respond to each point below and will incorporate the suggested analyses in the revision.

read point-by-point responses

  1. Referee: [Abstract and §4] Abstract and §4: the claim that rankings 'shift significantly across regimes' is presented without any statistical tests (e.g., rank-correlation p-values, permutation tests on rank differences, or bootstrap intervals), so it is impossible to determine whether the observed changes exceed sampling variability.

    Authors: We agree that formal statistical tests are required. In the revised manuscript we will add Kendall’s tau rank correlations (with p-values) between the model orderings obtained in each curvature regime. We will also report bootstrap confidence intervals on both per-model performance and on the rank positions themselves, computed over 1000 resamples of the test splits within each regime. These results will appear in Section 4 and will be referenced from the abstract. revision: yes

  2. Referee: [§3] §3 (Curvature Stratification) and methods: the partitioning into positive/negative/near-zero regimes depends on unspecified thresholds that are free parameters; no sensitivity analysis or justification is provided, so the reported within-regime stability and cross-regime shifts may be artifacts of these cutoffs.

    Authors: We accept that the thresholds must be justified and tested. The revision will explicitly state the curvature cut-offs (chosen at the 33rd and 66th percentiles of the empirical distribution of Ollivier-Ricci curvatures across the 14 datasets) together with a geometric motivation. We will add a sensitivity study that re-partitions the data at ±10 % and ±20 % shifts of these thresholds and confirms that the within-regime stability and cross-regime rank changes remain qualitatively unchanged. The sensitivity results will be placed in the supplementary material. revision: yes

  3. Referee: [§4] §4 (Experiments): no controls, matching, or multivariate regression are described for potential confounders (node/edge count, density, average degree, homophily, task type) that may covary with curvature regime; without such checks the attribution of ranking changes specifically to geometry rather than these covariates remains unverified.

    Authors: This is a valid methodological concern. The revised Section 4 will include multivariate regressions of model performance on curvature regime while controlling for the listed covariates. We will report both the coefficient on the curvature-regime indicator and the associated partial R² after accounting for the covariates. Variance-inflation factors will be checked, and, where feasible, we will apply propensity-score matching on the continuous covariates to compare performance across regimes on matched subsets. Any remaining limitations on isolating curvature will be discussed explicitly. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical benchmark with independent observations

full rationale

This paper presents an empirical benchmark study that partitions 14 datasets into curvature regimes (positive, negative, near-zero) and reports observed model performance rankings within and across strata. No equations, fitted parameters, or derivations are described that reduce to inputs by construction. The central claim rests on direct performance measurements rather than self-citation chains or ansatzes. Self-citations, if present, are not load-bearing for the reported results. This matches the default expectation of an honest non-finding for empirical work.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The framework rests on the domain assumption that curvature meaningfully stratifies relational data geometry; no free parameters or invented entities are described in the abstract.

free parameters (1)
  • curvature regime thresholds
    Boundaries separating positive, negative, and near-zero curvature regimes must be chosen or fitted to produce the three partitions.
axioms (1)
  • domain assumption Curvature is a computable intrinsic geometric property that governs relational model performance differences
    Invoked to justify stratification into regimes as the key latent factor.

pith-pipeline@v0.9.1-grok · 5817 in / 1158 out tokens · 47223 ms · 2026-06-28T02:37:00.161383+00:00 · methodology

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

62 extracted references · 4 linked inside Pith

  1. [1]

    Relational deep learning: Graph representation learning on relational databases.arXiv preprint arXiv:2312.04615, 2023

    Matthias Fey, Weihua Hu, Kexin Huang, Jan Eric Lenssen, Rishabh Ranjan, Joshua Robinson, Rex Ying, Jiaxuan You, and Jure Leskovec. Relational deep learning: Graph representation learning on relational databases.arXiv preprint arXiv:2312.04615, 2023

    arXiv 2023

  2. [2]

    Relbench: A benchmark for deep learning on relational databases

    Joshua Robinson, Rishabh Ranjan, Weihua Hu, Kexin Huang, Jiaqi Han, Alejandro Dobles, Matthias Fey, Jan Eric Lenssen, Yiwen Yuan, Zecheng Zhang, Xinwei He, and Jure Leskovec. Relbench: A benchmark for deep learning on relational databases. InAdvances in Neural Information Processing Systems, Datasets and Benchmarks Track, 2024

    2024

  3. [3]

    Kriege, Franka Bause, Kristian Kersting, Petra Mutzel, and Marion Neumann

    Christopher Morris, Nils M. Kriege, Franka Bause, Kristian Kersting, Petra Mutzel, and Marion Neumann. Tudataset: A collection of benchmark datasets for learning with graphs.arXiv preprint arXiv:2007.08663, 2020

    Pith/arXiv arXiv 2007

  4. [4]

    Feinberg, Joseph Gomes, Caleb Geniesse, Aneesh S

    Zhenqin Wu, Bharath Ramsundar, Evan N. Feinberg, Joseph Gomes, Caleb Geniesse, Aneesh S. Pappu, Karl Leswing, and Vijay Pande. Moleculenet: A benchmark for molecular machine learning.Chemical Science, 9(2):513–530, 2018

    2018

  5. [5]

    Carte: Pretraining and transfer for tabular learning

    Myung Jun Kim, Leo Grinsztajn, and Gael Varoquaux. Carte: Pretraining and transfer for tabular learning. InInternational Conference on Machine Learning, pages 23843–23866. PMLR, 2024

    2024

  6. [6]

    Bronstein, Joan Bruna, Taco Cohen, and Petar Veliˇckovi´c

    Michael M. Bronstein, Joan Bruna, Taco Cohen, and Petar Veliˇckovi´c. Geometric deep learning: Grids, groups, graphs, geodesics, and gauges.arXiv preprint arXiv:2104.13478, 2021

    Pith/arXiv arXiv 2021

  7. [7]

    Ricci curvature of markov chains on metric spaces.Journal of Functional Analysis, 256(3):810–864, 2009

    Yann Ollivier. Ricci curvature of markov chains on metric spaces.Journal of Functional Analysis, 256(3):810–864, 2009

    2009

  8. [8]

    Kipf and Max Welling

    Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. InInternational Conference on Learning Representations, 2017

    2017

  9. [9]

    Hamilton, Rex Ying, and Jure Leskovec

    William L. Hamilton, Rex Ying, and Jure Leskovec. Inductive representation learning on large graphs. InAdvances in Neural Information Processing Systems, 2017

    2017

  10. [10]

    Graph attention networks

    Petar Veliˇckovi´c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. Graph attention networks. InInternational Conference on Learning Representations, 2018

    2018

  11. [11]

    Kipf, Peter Bloem, Rianne van den Berg, Ivan Titov, and Max Welling

    Michael Schlichtkrull, Thomas N. Kipf, Peter Bloem, Rianne van den Berg, Ivan Titov, and Max Welling. Modeling relational data with graph convolutional networks. InEuropean Semantic Web Conference, pages 593–607. Springer, 2018

    2018

  12. [12]

    Graph foundation models: A comprehensive survey.arXiv preprint arXiv:2505.15116, 2025

    Zehong Wang, Zheyuan Liu, Tianyi Ma, Jiazheng Li, Zheyuan Zhang, Xingbo Fu, Yiyang Li, Zhengqing Yuan, Wei Song, Yijun Ma, Qingkai Zeng, Xiusi Chen, Jianan Zhao, Jundong Li, Meng Jiang, Pietro Liò, Nitesh Chawla, Chuxu Zhang, and Yanfang Ye. Graph foundation models: A comprehensive survey.arXiv preprint arXiv:2505.15116, 2025

    arXiv 2025

  13. [13]

    All in one and one for all: A simple yet effective method towards cross-domain graph pretraining

    Haihong Zhao, Aochuan Chen, Xiangguo Sun, Hong Cheng, and Jia Li. All in one and one for all: A simple yet effective method towards cross-domain graph pretraining. InProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2024

    2024

  14. [14]

    Text-free multi-domain graph pre-training: Toward graph foundation models.arXiv preprint arXiv:2405.13934, 2024

    Xingtong Yu, Chang Zhou, Yuan Fang, and Xinming Zhang. Text-free multi-domain graph pre-training: Toward graph foundation models.arXiv preprint arXiv:2405.13934, 2024

    arXiv 2024

  15. [15]

    Multi-domain graph foundation models: Robust knowledge transfer via topology alignment

    Shuo Wang, Bokui Wang, Zhixiang Shen, Boyan Deng, and Zhao Kang. Multi-domain graph foundation models: Robust knowledge transfer via topology alignment. InInternational Conference on Machine Learning, 2025

    2025

  16. [16]

    Poincaré embeddings for learning hierarchical represen- tations

    Maximilian Nickel and Douwe Kiela. Poincaré embeddings for learning hierarchical represen- tations. InAdvances in Neural Information Processing Systems, 2017

    2017

  17. [17]

    Hyperbolic graph convolutional neural networks

    Ines Chami, Zhitao Ying, Christopher Ré, and Jure Leskovec. Hyperbolic graph convolutional neural networks. InAdvances in Neural Information Processing Systems, 2019. 11

    2019

  18. [18]

    Design space for graph neural networks

    Jiaxuan You, Rex Ying, and Jure Leskovec. Design space for graph neural networks. In Advances in Neural Information Processing Systems, 2020

    2020

  19. [19]

    Open graph benchmark: Datasets for machine learning on graphs

    Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta, and Jure Leskovec. Open graph benchmark: Datasets for machine learning on graphs. InAdvances in Neural Information Processing Systems, 2020

    2020

  20. [20]

    Joshi, Anh Tuan Luu, Thomas Laurent, Yoshua Bengio, and Xavier Bresson

    Vijay Prakash Dwivedi, Chaitanya K. Joshi, Anh Tuan Luu, Thomas Laurent, Yoshua Bengio, and Xavier Bresson. Benchmarking graph neural networks.Journal of Machine Learning Research, 24(43):1–48, 2023

    2023

  21. [21]

    Bochner’s method for cell complexes and combinatorial ricci curvature.Discrete & Computational Geometry, 29(3):323–374, 2003

    Robin Forman. Bochner’s method for cell complexes and combinatorial ricci curvature.Discrete & Computational Geometry, 29(3):323–374, 2003

    2003

  22. [22]

    Ricci curvature of the internet topology

    Chien-Chun Ni, Yu-Yao Lin, Jie Gao, Xianfeng David Gu, and Emil Saucan. Ricci curvature of the internet topology. InIEEE Conference on Computer Communications, pages 2758–2766. IEEE, 2015

    2015

  23. [23]

    Hyperbolic neural networks

    Octavian-Eugen Ganea, Gary Bécigneul, and Thomas Hofmann. Hyperbolic neural networks. InAdvances in Neural Information Processing Systems, 2018

    2018

  24. [24]

    Hyperbolic graph neural networks

    Qi Liu, Maximilian Nickel, and Douwe Kiela. Hyperbolic graph neural networks. InAdvances in Neural Information Processing Systems, 2019

    2019

  25. [25]

    Mixed-curvature variational autoencoders

    Ondrej Skopek, Octavian-Eugen Ganea, and Gary Bécigneul. Mixed-curvature variational autoencoders. InInternational Conference on Learning Representations, 2020

    2020

  26. [26]

    Constant curvature graph convolutional networks.Proceedings of Machine Learning Research, 119:486–496, 2020

    Gregor Bachmann, Gary Bécigneul, and Octavian-Eugen Ganea. Constant curvature graph convolutional networks.Proceedings of Machine Learning Research, 119:486–496, 2020

    2020

  27. [27]

    Learning mixed-curvature rep- resentations in product spaces

    Albert Gu, Frederic Sala, Beliz Gunel, and Christopher Ré. Learning mixed-curvature rep- resentations in product spaces. InInternational Conference on Learning Representations, 2019

    2019

  28. [28]

    Pseudo- riemannian graph convolutional networks

    Bo Xiong, Shichao Zhu, Nico Potyka, Shirui Pan, Chuan Zhou, and Steffen Staab. Pseudo- riemannian graph convolutional networks. InAdvances in Neural Information Processing Systems, 2022

    2022

  29. [29]

    Ioannidis, and Christos Faloutsos

    Karish Grover, Haiyang Yu, Xiang Song, Qi Zhu, Han Xie, Vassilis N. Ioannidis, and Christos Faloutsos. Spectro-riemannian graph neural networks. InInternational Conference on Learning Representations, 2025

    2025

  30. [30]

    American Mathematical Society, Providence, RI, 2001

    Dmitri Burago, Yuri Burago, and Sergei Ivanov.A Course in Metric Geometry, volume 33 of Graduate Studies in Mathematics. American Mathematical Society, Providence, RI, 2001

    2001

  31. [31]

    Bridson and André Haefliger.Metric Spaces of Non-Positive Curvature, volume 319 ofGrundlehren der mathematischen Wissenschaften

    Martin R. Bridson and André Haefliger.Metric Spaces of Non-Positive Curvature, volume 319 ofGrundlehren der mathematischen Wissenschaften. Springer, Berlin, Heidelberg, 1999

    1999

  32. [32]

    D. N. Joanes and C. A. Gill. Comparing measures of sample skewness and kurtosis.Journal of the Royal Statistical Society: Series D (The Statistician), 47(1):183–189, 1998

    1998

  33. [33]

    Revisiting semi-supervised learning with graph embeddings.arXiv (Cornell University), 2022

    Zhilin Yang, William Cohen, and Ruslan Salakhutdinov. Revisiting semi-supervised learning with graph embeddings.arXiv (Cornell University), 2022

    2022

  34. [34]

    Geom-gcn: Geometric graph convolutional networks.ICLR, 2020

    Hongbin Pei, Bingzhe Wei, Kevin Chang, Yu Lei, and Bo Yang. Geom-gcn: Geometric graph convolutional networks.ICLR, 2020

    2020

  35. [35]

    Telegraph: A benchmark dataset for hierarchical link prediction.arXiv (Cornell University), abs/2204.07703, 2022

    Min Zhou, Bisheng Li, Menglin Yang, and Lujia Pan. Telegraph: A benchmark dataset for hierarchical link prediction.arXiv (Cornell University), abs/2204.07703, 2022

    arXiv 2022

  36. [36]

    Rossi and Nesreen K

    Ryan A. Rossi and Nesreen K. Ahmed. The network data repository with interactive graph analytics and visualization. InAAAI, 2015. 12

    2015

  37. [37]

    The predictive toxicology evaluation challenge

    SH Muggleton. The predictive toxicology evaluation challenge. InIJCAI-97: Proceedings of the Fifteenth International Joint Conference on Artificial Intelligence, Nagoya, Japan, August 23-29, 1997, volume 2, page 4. Morgan Kaufmann, 1997

    1997

  38. [38]

    Policy-based memoization for ilp-based concept discovery systems.Journal of Intelligent Information Systems, 46(1):99–120, 2016

    Alev Mutlu and Pinar Karagoz. Policy-based memoization for ilp-based concept discovery systems.Journal of Intelligent Information Systems, 46(1):99–120, 2016

    2016

  39. [39]

    Pc-conv: Unifying homophily and heterophily with two-fold filtering

    Bingheng Li, Erlin Pan, and Zhao Kang. Pc-conv: Unifying homophily and heterophily with two-fold filtering. InProceedings of the AAAI conference on artificial intelligence, volume 38, pages 13437–13445, 2024

    2024

  40. [40]

    Hyperbolic graph attention network.IEEE Transactions on Big Data, 8(6):1690–1701, 2021

    Yiding Zhang, Xiao Wang, Chuan Shi, Xunqiang Jiang, and Yanfang Ye. Hyperbolic graph attention network.IEEE Transactions on Big Data, 8(6):1690–1701, 2021

    2021

  41. [41]

    Fully hyperbolic neural networks

    Weize Chen, Xu Han, Yankai Lin, Hexu Zhao, Zhiyuan Liu, Peng Li, Maosong Sun, and Jie Zhou. Fully hyperbolic neural networks. InProceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 5672–5686, 2022

    2022

  42. [42]

    Graphmore: Mitigating topological heterogeneity via mixture of riemannian experts

    Zihao Guo, Qingyun Sun, Haonan Yuan, Xingcheng Fu, Min Zhou, Yisen Gao, and Jianxin Li. Graphmore: Mitigating topological heterogeneity via mixture of riemannian experts. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 11754–11762, 2025

    2025

  43. [43]

    Samgpt: Text-free graph foundation model for multi-domain pre-training and cross-domain adaptation

    Xingtong Yu, Zechuan Gong, Chang Zhou, Yuan Fang, and Hui Zhang. Samgpt: Text-free graph foundation model for multi-domain pre-training and cross-domain adaptation. InProceedings of the ACM Web Conference, 2025

    2025

  44. [44]

    Li Sun, Zhenhao Huang, Silei Chen, Lanxu Yang, Junda Ye, Sen Su, and Philip S. Yu. Multi- domain riemannian graph gluing for building graph foundation models. InInternational Conference on Learning Representations, 2026

    2026

  45. [45]

    Sa2gfm: Enhancing robust graph foundation models with structure-aware semantic augmentation

    Junhua Shi, Qingyun Sun, Haonan Yuan, and Xingcheng Fu. Sa2gfm: Enhancing robust graph foundation models with structure-aware semantic augmentation. InProceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 15716–15724, 2026

    2026

  46. [46]

    On lipschitz embedding of finite metric spaces in hilbert space.Israel Journal of Mathematics, 52(1-2):46–52, 1985

    Jean Bourgain. On lipschitz embedding of finite metric spaces in hilbert space.Israel Journal of Mathematics, 52(1-2):46–52, 1985

    1985

  47. [47]

    The geometry of graphs and some of its algorithmic applications.Combinatorica, 15(2):215–245, 1995

    Nathan Linial, Eran London, and Yuri Rabinovich. The geometry of graphs and some of its algorithmic applications.Combinatorica, 15(2):215–245, 1995

    1995

  48. [48]

    The proof and measurement of association between two things.The American Journal of Psychology, 15(1):72–101, 1904

    Charles Spearman. The proof and measurement of association between two things.The American Journal of Psychology, 15(1):72–101, 1904

    1904

  49. [49]

    Maurice G. Kendall. A new measure of rank correlation.Biometrika, 30(1/2):81–93, 1938

    1938

  50. [50]

    Étude comparative de la distribution florale dans une portion des alpes et du jura

    Paul Jaccard. Étude comparative de la distribution florale dans une portion des alpes et du jura. Bulletin de la Société Vaudoise des Sciences Naturelles, 37:547–579, 1901

    1901

  51. [51]

    Lorentzian graph convolu- tional networks

    Yiding Zhang, Xiao Wang, Chuan Shi, Nian Liu, and Guojie Song. Lorentzian graph convolu- tional networks. InProceedings of The Web Conference, pages 1249–1261, 2021

    2021

  52. [52]

    Curvature graph neural network.Information Sciences, 592:50–66, 2022

    Haifeng Li, Jun Cao, Jiawei Zhu, Yu Liu, Qing Zhu, and Guohua Wu. Curvature graph neural network.Information Sciences, 592:50–66, 2022

    2022

  53. [53]

    Position: Graph foundation models are already here

    Haitao Mao, Zhikai Chen, Wenzhuo Tang, Jianan Zhao, Yao Ma, Tong Zhao, Neil Shah, Mikhail Galkin, and Jiliang Tang. Position: Graph foundation models are already here. InProceedings of the 41st International Conference on Machine Learning, 2024

    2024

  54. [54]

    Relational inductive biases, deep learning, and graph networks.arXiv preprint arXiv:1806.01261, 2018

    Peter W Battaglia, Jessica B Hamrick, Victor Bapst, Alvaro Sanchez-Gonzalez, Vinicius Zambaldi, Mateusz Malinowski, Andrea Tacchetti, David Raposo, Adam Santoro, Ryan Faulkner, et al. Relational inductive biases, deep learning, and graph networks.arXiv preprint arXiv:1806.01261, 2018. 13

    Pith/arXiv arXiv 2018

  55. [55]

    A survey on foundation models for structured data: Tabular, time series, and graphs

    Qingyun Sun, Haonan Yuan, Yi Huang, Ziwei Zhang, Xingcheng Fu, Ruijie Wang, Haoyi Zhou, Jia Wu, Jianxin Li, and Philip S Yu. A survey on foundation models for structured data: Tabular, time series, and graphs. 2026

    2026

  56. [56]

    Turning tabular foundation models into graph foundation models.arXiv preprint arXiv:2508.20906, 2025

    Dmitry Eremeev, Gleb Bazhenov, Oleg Platonov, Artem Babenko, and Liudmila Prokhorenkova. Turning tabular foundation models into graph foundation models.arXiv preprint arXiv:2508.20906, 2025

    Pith/arXiv arXiv 2025

  57. [57]

    Graphoracle: Efficient fully-inductive knowledge graph reasoning via relation-dependency graphs

    Enjun Du, Siyi Liu, and Yongqi Zhang. Graphoracle: Efficient fully-inductive knowledge graph reasoning via relation-dependency graphs. InProceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 19055–19063, 2026

    2026

  58. [58]

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

    Qingsong Lv, Ming Ding, Qiang Liu, Yuxiang Chen, Wenzheng Feng, Siming He, Chang Zhou, Jianguo Jiang, Yuxiao Dong, and Jie Tang. Are we really making much progress? revisiting, benchmarking and refining heterogeneous graph neural networks. InProceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining, pages 1150–1160, 2021

    2021

  59. [59]

    Graph neural network: A comprehensive review on non-euclidean space.Ieee Access, 9:60588–60606, 2021

    Nurul A Asif, Yeahia Sarker, Ripon K Chakrabortty, Michael J Ryan, Md Hafiz Ahamed, Dip K Saha, Faisal R Badal, Sajal K Das, Md Firoz Ali, Sumaya I Moyeen, et al. Graph neural network: A comprehensive review on non-euclidean space.Ieee Access, 9:60588–60606, 2021

    2021

  60. [60]

    Mose: Unveiling structural patterns in graphs via mixture of subgraph experts.arXiv preprint arXiv:2509.09337, 2025

    Junda Ye, Zhongbao Zhang, Li Sun, and Siqiang Luo. Mose: Unveiling structural patterns in graphs via mixture of subgraph experts.arXiv preprint arXiv:2509.09337, 2025

    arXiv 2025

  61. [61]

    Cooperation of experts: Fusing heterogeneous information with large margin

    Shuo Wang, Shunyang Huang, Jinghui Yuan, Zhixiang Shen, and Zhao Kang. Cooperation of experts: Fusing heterogeneous information with large margin. InInternational Conference on Machine Learning, pages 63169–63185. PMLR, 2025

    2025

  62. [62]

    fatter”-than-Euclidean geometry, while negative values indicate locally “thinner

    Zihao Guo, Qingyun Sun, Haonan Yuan, Xingcheng Fu, Min Zhou, Yisen Gao, and Jianxin Li. Graphmore: Mitigating topological heterogeneity via mixture of riemannian experts. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 11754–11762, 2025. 14 Contents 1 Introduction 1 2 Graph Curvature 2 3 The Setup of CURVBENCH3 3.1 Datas...

    2025