RiemannGFM: Learning a Graph Foundation Model from Riemannian Geometry

Hao Peng; Li Sun; Philip Yu; Qiqi Wan; Suyang Zhou; Zhenhao Huang

REVIEW 4 major objections 4 minor 1 cited by

RiemannGFM: Learning a Graph Foundation Model from Riemannian Geometry

T0 review · 4 major / 4 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read RiemannGFM claims that a single model pretrained only on the structural vocabulary of trees and cycles, embedded in appropriately curved Riemannian spaces, can transfer across arbitrary graphs—including graphs with no textual attributes.

desk verdict A genuinely novel structure-only pretraining architecture whose 'any graph' claim currently rests on unaligned spectral inputs and a shaky geometric proof; worth reviewing but not ready as stated. read the letter →

arxiv 2502.03251 v2 pith:FY6TMZ2T submitted 2025-02-05 cs.LG

Li Sun , Zhenhao Huang , Suyang Zhou , Qiqi Wan , Hao Peng , Philip Yu This is my paper · ORCID

classification cs.LG

keywords graphfoundationmodelRiemanniangeometrystructuralvocabularyconstantcurvaturespacehyperbolichypersphericalproductbundlecross-domaintransferlearning

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that a graph foundation model does not need text attributes or language models to transfer across domains: the shared 'vocabulary' of trees and cycles is enough. The proposed model, RiemannGFM, learns that vocabulary in curved geometric spaces, placing trees in hyperbolic space and cycles in hyperspherical space, and is pretrained with a geometric contrastive objective. If the claim is right, one pretrained structure-only encoder can serve arbitrary graphs, which would matter for the many real graphs that carry no useful text. The paper supports the claim with cross-domain link prediction and few-shot node classification experiments, including a Citeseer link-prediction AUC of 99.40 against 96.90 for a strong self-supervised baseline.

What carries the argument

The load-bearing object is the product bundle: a manifold paired with its tangent spaces, formed as the Cartesian product of a hyperbolic factor (for trees) and a hyperspherical factor (for cycles), written $\mathcal{P}^{d_P} = (\mathbb{H}^{d_H}_{\kappa_H}\otimes T\mathbb{H}^{d_H}_{\kappa_H})\otimes(\mathbb{S}^{d_S}_{\kappa_S}\otimes T\mathbb{S}^{d_S}_{\kappa_S})$. Node coordinates on the manifolds carry local substructure geometry, while node encodings in the tangent spaces carry global structure; bundle convolution resolves the incompatibility of different tangent spaces by parallel transport, and the geometric midpoint aggregates coordinates across sampled substructures. The vocabulary learning module uses a manifold-preserving linear operation and cross-geometry attention (query from one geometry, key and value from the other), and the geometric contrastive loss contrasts the two views in the shared tangent space of the north pole.

What would settle it

A concrete test: take the frozen pretrained RiemannGFM and, for each target graph, randomly flip the signs or permute the rows of its top-$K$ Laplacian eigenvectors before generating encodings. If link-prediction AUC on Citeseer or Airport stays near the reported values, the eigenvectors are not the carrier of transfer; if it collapses, the method depends on a per-graph spectral alignment that the paper does not provide.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the graph domain has a structural vocabulary—trees and cycles—and that this vocabulary maps onto constant-curvature Riemannian geometry: trees align with hyperbolic space, cycles with hyperspherical space. The model pairs each node with a coordinate on a manifold (its position inside a sampled substructure) and an encoding in the tangent space at that coordinate (its global-structure information), combined into a product bundle $\mathcal{P}^{d_P} = (\mathbb{H}^{d_H}_{\kappa_H}\otimes T\mathbb{H}^{d_H}_{\kappa_H})\otimes(\mathbb{S}^{d_S}_{\kappa_S}\otimes T\mathbb{S}^{d_S}_{\kappa_S})$. A universal Riemannian layer alternates cross-geometry attention, which updates coordinates within substructures, and bundle convolution, which transports encodings between tangent spaces via parallel transport. Pretraining is self-supervised through a contrastive loss between the hyperbolic and hyperspherical views, with only the top-$K$ eigenvectors of the normalized graph Laplacian as input; the same frozen parameters then generate node encodings for an unseen graph.

Load-bearing premise

The entire pretraining signal rests on the assumption that the top few eigenvectors of a graph's normalized Laplacian—computed independently for each graph, with arbitrary signs and orderings—encode a reusable structural vocabulary that transfers across very different graphs.

Editorial extensions

If this is right

A graph foundation model can be built without text: structure alone transfers across citation, social, collaboration, and airport networks, so text-free graphs are no longer excluded from pretraining.
The same frozen encoder competes with specialized models on link prediction; on Citeseer it reaches 99.40 AUC, above DGI's 96.90, and it leads the reported baselines on Pubmed, GitHub, and Airport.
Few-shot node classification improves most on non-attributed graphs, for example 77.83 accuracy in the 1-shot setting and 84.19 accuracy in the 5-shot setting on GitHub, where text-based models show negative transfer.
The geometry pairing is doing real work: trees in hyperbolic space and cycles in hyperspherical space beat the Euclidean, all-hyperbolic, and all-spherical alternatives in the paper's ablations.
Because the only input is a fixed-size set of Laplacian eigenvectors, the model applies to any graph regardless of node feature dimension, and it supports minibatch training.

Reading between the lines

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

(Editorial inference) If eigenvectors are truly the only input, the paper implies that graph structure alone carries a universal signal; a natural next test is whether random or sign-flipped eigenvectors destroy transfer, which would reveal how much of the result depends on exact spectral alignment.
(Editorial inference) The tree-and-cycle vocabulary is minimal, and the same product-bundle design can accommodate additional substructures or curvatures, so the framework may extend to motif-rich graphs beyond the two-geometry case.
(Editorial inference) The geometric contrastive loss can be read as a parameter-free form of multi-view contrastive learning, suggesting a bridge between geometry-aware self-supervision and standard graph contrastive methods that rely on augmentation.
(Editorial inference) A practical consequence the authors do not spell out is that the method applies to graphs whose node features are missing or privacy-sensitive, since node identity is represented purely through spectral structure.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Colleague,

The genuinely new thing here is the combination: a structure-only graph pretraining model whose shared vocabulary is trees and cycles, embedded in a product of hyperbolic and hyperspherical bundles with tangent-space encodings. I don't know another GFM that does this without text. The paper ships code, gives ablations for the geometry choices, and shows impressive link prediction transfer on non-text graphs (Airports, GitHub). Credit where due: the architecture is inventive and the empirical wins are at least plausible.

The soft spots are real, though. The "any graph" claim rests on top-K eigenvectors of the normalized graph Laplacian as the only input. Those eigenvectors are defined only up to sign, and up to basis rotation in degenerate eigenspaces. The paper normalizes only by fixing K; there is no sign alignment, no sign-invariant encoding, no sign augmentation. Since the attention and bundle convolution are not sign-invariant functions of their inputs, a global sign flip changes the encodings. Pretraining on three source graphs uses one arbitrary sign convention per graph; target graphs can carry different ones. The paper never checks this. That is a load-bearing unknown, not a nit. Also, using the largest eigenvectors of the normalized Laplacian is unusual—I'd want to know whether they meant the smallest.

Second, Theorem 2's proof identifies squared distance with 2/κ − 2⟨x,y⟩_κ, which is the chordal distance, not the squared geodesic distance on the sphere or hyperboloid. So the "geometric midpoint" is not the Fréchet mean it is claimed to be. The aggregation may still work in practice, but the theorem as stated does not follow from that proof.

Third, Table 1 has no error bars or significance tests, and the downstream heads are under-specified. Those are fixable, but they matter for the headline transfer numbers. The universality claim also rests on only four target datasets, which is thin for "any graph."

The central idea is good enough to deserve referee time. I'd send it to review, but with a clear request: handle or justify the eigenvector sign/basis issue, correct or downgrade the midpoint theorem, and add variance and error bars to the main table. The paper is for graph-ML readers working on pretraining and geometric GNNs; it does not reorganize a field, but it is a step beyond text-coupled GFMs.

Referee Report

4 major / 4 minor

Summary. The paper proposes RiemannGFM, a self-supervised graph foundation model that uses only structural information, without textual attributes. The model is built around a 'structural vocabulary' of trees and cycles, which are respectively embedded in hyperbolic and spherical spaces via a product bundle that pairs manifold coordinates with tangent-space node encodings. The architecture consists of a cross-geometry attention module, a bundle convolution based on parallel transport, and a geometric contrastive loss between hyperbolic and spherical views. The model is pretrained on ogbn-arxiv, Physics, and Amazon-Computers and evaluated on Citeseer, Pubmed, GitHub, and Airports for node classification and link prediction, with additional few-shot, ablation, and pretraining-dataset studies. The central claim is that a single structure-only model can transfer across arbitrary graph domains by learning shared geometric structure.

Significance. If the claims hold, RiemannGFM would be a genuinely novel direction for graph foundation models: it moves beyond LLM-based, text-attributed approaches and provides a structure-only pretraining paradigm grounded in Riemannian geometry. The paper deserves credit for proposing a concrete structural-vocabulary hypothesis (trees to hyperbolic, cycles to spherical), releasing code, and including ablations over geometry choices and pretraining datasets. However, the significance is conditional on two load-bearing points: whether unaligned Laplacian eigenvectors can serve as a transferable input representation, and whether the geometric midpoint operation is mathematically correct as stated. The current manuscript does not establish either point, and the main experimental table lacks error bars, so the empirical support is weaker than the text suggests.

major comments (4)

[Sec. 4.1.3 and App. E.3.2] The model's only input is the top-K eigenvectors of the normalized graph Laplacian, but the paper neither aligns eigenvector signs/bases across graphs nor uses any sign-invariant mechanism. Since eigenvectors are defined only up to sign (and, for degenerate eigenspaces, up to arbitrary orthogonal rotations), a global sign flip of the input will generally change the outputs of the attention in Eq. (6) and the bundle convolution in Eq. (7), because these operations are not sign-invariant. Pretraining on ogbn-arxiv, Physics, and Amazon-Computers therefore fixes one arbitrary sign convention per dataset, while target graphs may carry different conventions, so the reported cross-domain transfer could be an artifact of accidentally consistent sign choices. The authors should test this directly: for example, apply random sign flips (and random basis rotations within degenerate eigenspaces) to the input eigenvectors at evaluation time and report the variance in downstream metrics, or train with sign-flip augmentation and verify robustness. Alternatively, they should replace eigenvector inputs with sign-invariant spectral features. This issue is load-bearing because the universality claim rests entirely on the transferability of these input features.
[Sec. 3.1.2 and App. B.2, Eq. (4) and Eq. (5)] Theorem 2 is not proved as stated. The proof identifies the squared distance with the chordal expression 2/κ − 2⟨x, y⟩_κ, but this is not the squared geodesic distance on the hyperboloid or sphere. The weighted arithmetic mean in Eq. (4) is the normalized projection of a weighted average in the ambient space, and it minimizes the sum of squared chordal distances, not the sum of squared geodesic distances. Therefore the claim that Eq. (4) is the geometric midpoint (Fréchet mean) with respect to geodesic distance is incorrect, and Eq. (5) does not implement the stated geometric midpoint. This matters because the vocabulary learning module and graph-level alignment both rely on the midpoint operation. The authors should either prove the result for the true geodesic distance, replace the claim with a chordal-distance centroid and adjust the narrative, or implement a Fréchet mean and re-derive the attention update accordingly.
[Sec. 3.1.2, Eq. (3) and Theorem 1] The manifold-preserving linear operation in Eq. (3) is undefined whenever W x_s = 0, since the re-scaling factor α divides by ||W x_s||. This includes the north pole x = [1/sqrt(|κ|), 0, ..., 0] for any W, so Theorem 1's claim 'for any W' is not literally true. The authors should state the domain where the operation is defined and explain how the model handles nodes whose spatial coordinate is zero or whose transformed spatial coordinate vanishes, especially during initialization from zero input encodings.
[Sec. 4.2.1, Table 1, and App. E] The main transfer-learning table reports no standard deviations or significance tests, and the downstream protocol for link prediction is under-specified. In particular, the text claims 'consistently achieves the best results' for link prediction, but some node-classification gaps are small (e.g., GitHub ACC 85.96 vs 85.68 for GCN; Airport ACC 55.29 vs 52.34 for GraphMAE2), and without error bars or significance tests these differences are not evaluable. Furthermore, the paper does not describe how link prediction heads are trained or evaluated on new graphs (negative sampling, train/val/test split, frozen vs fine-tuned encodings), nor how featureless baselines (GCN, SAGE, DGI, GraphMAE2) are applied to GitHub and Airports, which have feature dimension 0 in Table 6. The authors should add error bars/significance tests to Table 1 and fully specify the evaluation protocols in Appendix E.

minor comments (4)

[Throughout] There are several typos and citation inconsistencies: 'RemannGFM' in the conclusion, 'prformance' in the Table 4 caption, 'eignvectors' in App. E.3.2, and 'Importation Notations' in Table 5. In Table 3, OpenGraph is cited as [59] but the correct reference appears to be [50].
[Eq. (2)] The metric of the product bundle is written with I_{d_H+1} for both the hyperbolic and spherical tangent factors; the second identity should likely be I_{d_S+1}.
[App. B.3] The definition of the curvature-aware cosine says 'cos_κ(·) = cosh(·) with κ > 0', but for the hyperbolic case one would expect κ < 0; this appears to be a typo.
[App. B.2] The proof of the midpoint switches between indices i and j (e.g., 'Σ_i ν_i x_j' and 'Σ_{j∈N_i} ν_{ij} h_j'), which makes the derivation hard to follow; the notation should be cleaned up.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the tree/cycle vocabulary and hyperbolic/spherical pairing rest on external results, and transfer is measured against external benchmarks.

full rationale

The paper's core chain is not circular. The structural vocabulary of trees and cycles is justified by the elementary fact that any connected graph has a spanning tree and that each additional edge creates a cycle, and the pairing of trees with hyperbolic space and cycles with hyperspherical space is supported by external results (Sarkar's low-distortion tree embedding; rotational invariance from a Riemannian geometry textbook), not by parameters fitted to the target outcome. Node encodings are initialized from the top-K eigenvectors of the normalized graph Laplacian (Sec. 4.1.3, App. E.3.2), which is an input representation independent of the transfer labels; the pretraining contrastive loss is self-supervised, and the reported cross-domain and few-shot numbers are compared against strong external baselines (GCN, SAGE, DGI, GraphMAE2, OFA, GCOPE, LLaGA, OpenGraph). No load-bearing claim is justified by a self-citation chain: the authors' prior work appears only in related-work context. One mathematical derivation, Theorem 2/Appendix B.2, uses the identity d^2 = 2/κ − 2⟨x,y⟩_κ, which is not the squared geodesic distance on the Lorentz/sphere model; this is a correctness concern about the 'geometric midpoint' result, but it is not an input-output reduction and does not make the empirical transfer claim circular. Overall, the central derivation is self-contained against external evidence, so circularity burden is low.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central product-bundle construction rests on four classes of inputs: a graph-theoretic decomposition into trees and cycles, prior curvature-structure alignment results, an unproved geometric-mean claim, and the Laplacian-eigenvector input representation. No fitted constants or new physical entities are introduced; the main free choices are hyperparameters.

free parameters (3)

Eigenvector count K for Laplacian input = not reported
Node encodings are initialized from the top-K eigenvectors of the normalized graph Laplacian (Appendix E.3.2). K is called predefined but never given a value, and it controls the structural signal fed to all layers.
Product bundle factor dimensions and curvatures = d_H = d_S = 32; kappa_H = -1, kappa_S = 1
Set by grid search (Appendix E.3.3). These hyperparameters determine capacity and geometry and are not derived from the problem.
Architecture hyperparameters = 2 layers, hidden dimension 256, dropout 0.1, learning rate 0.01
Tuned with grid search. They affect transfer performance but are conventional choices rather than fitted constants.

assumptions (5)

standard math Trees and cycles form a structural vocabulary that can construct arbitrary graphs (Definition 1).
Any connected graph has a spanning tree plus extra edges that form cycles; the paper uses this to justify sampling only trees and cycles (Sec 3).
domain assumption Hyperbolic space aligns with tree structures and hyperspherical space aligns with cycles.
The geometry choice rests on cited results, Sarkar's bounded-distortion tree embedding [31] and rotational invariance of cycles [28], not on a proof in this paper (Sec 3.1.1).
ad hoc to paper The weighted arithmetic mean in Eq. (4) is the geometric midpoint with respect to squared distance.
Theorem 2 is central to node coordinate updates. The proof uses d^2 = 2/kappa - 2<x,y>_kappa, which is chordal distance on the sphere, so the geodesic-midpoint claim is not established for the spherical factor (Appendix B.2).
domain assumption Top-K Laplacian eigenvectors provide a universal structural input that transfers across graphs.
Node encodings are initialized from eigenvectors of the normalized Laplacian, with no alignment or sign handling across datasets (Sec 4.1.3, Appendix E.3.2).
standard math Parallel transport along geodesics in Lorentz/Spherical models is given by Eq. (20).
Standard Riemannian geometry result referenced to Petersen [28]; used in bundle convolution and contrastive views.

how reviews work

0 comments

Cite this review

Pith. "Pith review of RiemannGFM: Learning a Graph Foundation Model from Riemannian Geometry." pith.science (2026). https://pith.science/paper/FY6TMZ2T

@misc{pith2026250203251,
  author       = {Pith},
  title        = {Pith review of: RiemannGFM: Learning a Graph Foundation Model from Riemannian Geometry},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FY6TMZ2T}},
  note         = {Machine review of arXiv:2502.03251}
}

read the original abstract

The foundation model has heralded a new era in artificial intelligence, pretraining a single model to offer cross-domain transferability on different datasets. Graph neural networks excel at learning graph data, the omnipresent non-Euclidean structure, but often lack the generalization capacity. Hence, graph foundation model is drawing increasing attention, and recent efforts have been made to leverage Large Language Models. On the one hand, existing studies primarily focus on text-attributed graphs, while a wider range of real graphs do not contain fruitful textual attributes. On the other hand, the sequential graph description tailored for the Large Language Model neglects the structural complexity, which is a predominant characteristic of the graph. Such limitations motivate an important question: Can we go beyond Large Language Models, and pretrain a universal model to learn the structural knowledge for any graph? The answer in the language or vision domain is a shared vocabulary. We observe the fact that there also exist shared substructures underlying graph domain, and thereby open a new opportunity of graph foundation model with structural vocabulary. The key innovation is the discovery of a simple yet effective structural vocabulary of trees and cycles, and we explore its inherent connection to Riemannian geometry. Herein, we present a universal pretraining model, RiemannGFM. Concretely, we first construct a novel product bundle to incorporate the diverse geometries of the vocabulary. Then, on this constructed space, we stack Riemannian layers where the structural vocabulary, regardless of specific graph, is learned in Riemannian manifold offering cross-domain transferability. Extensive experiments show the effectiveness of RiemannGFM on a diversity of real graphs.

Figures

Figures reproduced from arXiv: 2502.03251 by the authors.

**Figure 2.** An illustration of bundle convolution. the corresponding Riemannian manifold with cross-geometry keys, while Hypformer places the input as a whole in hyperbolic space. 3.1.4 Global Learning Module. Sampling multiple substructures from the graph, this module examines the entire graph to learn node encodings from a global perspective. This objective is achieved by the following two phases. Firstly, we study the node … view at source ↗

**Figure 4.** Link prediction results with structural knowledge [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

Figures from the paper (1 more)

**Figure 5.** Figure 5: Visualization on Cora for analyzing graphs. LLaGA [5] tailors graphs for the language model with node sequences, generated via graph translation, while OFA [21] unifies different graph data by the language description of nodes and edges. OpenGraph [50] re-frames textua…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

Towards Non-Euclidean Foundation Models: Advancing AI Beyond Euclidean Frameworks
cs.CG 2025-05 unverdicted novelty 2.0 of 10

A workshop proposal outlining the case for combining non-Euclidean geometry with foundation models for web applications.

Reference graph

Works this paper leans on

67 extracted references · 59 canonical work pages · cited by 1 Pith paper

[1]

Gregor Bachmann, Gary Bécigneul, and Octavian Ganea. 2020. Constant Curva- ture Graph Convolutional Networks. In Proceedings of the 37th ICML , Vol. 119. PMLR, 486–496

work page 2020
[2]

Dominique Beaini, Shenyang Huang, and Joao Alex Cunha et. al. 2024. Towards Foundational Models for Molecular Learning on Large-Scale Multi-Task Datasets. In Proceedings of the 12th ICLR . OpenReview.net

work page 2024
[3]

Ines Chami, Zhitao Ying, Christopher Ré, and Jure Leskovec. 2019. Hyperbolic Graph Convolutional Neural Networks. In Advances in the 32nd NeurIPS . 4869– 4880

work page 2019
[4]

Yupeng Chang, Xu Wang, and Jindong Wang. 2024. A Survey on Evaluation of Large Language Models. ACM Trans. Intell. Syst. Technol. 15, 3 (2024), 39:1–39:45

work page 2024
[5]

Runjin Chen, Tong Zhao, Ajay Kumar Jaiswal, Neil Shah, and Zhangyang Wang
[6]

Weize Chen, Xu Han, Yankai Lin, Hexu Zhao, Zhiyuan Liu, Peng Li, Maosong Sun, and Jie Zhou. 2022. Fully Hyperbolic Neural Networks. In Proceedings of the 60th ACL. ACL, 5672–5686

work page 2022
[7]

Haitz Sáez de Ocáriz Borde, Anees Kazi, Federico Barbero, and Pietro Liò. 2023. Latent Graph Inference using Product Manifolds. In Proceedings of the 11th ICLR . OpenReview.net

work page 2023
[8]

Michaël Defferrard, Xavier Bresson, and Pierre Vandergheynst. 2016. Convo- lutional Neural Networks on Graphs with Fast Localized Spectral Filtering. In Advances in the 29th NeurIPS . 3837–3845

work page 2016

Show all 67 references

[9]

Mikhail Galkin, Xinyu Yuan, Hesham Mostafa, Jian Tang, and Zhaocheng Zhu
[10]

Octavian-Eugen Ganea, Gary Bécigneul, and Thomas Hofmann. 2018. Hyperbolic Neural Networks. In Advances in the 31st NeurIPS . 5350–5360

2018
[11]

In Proceed- ings of the 12th ICLR

Towards Foundation Models for Knowledge Graph Reasoning. In Proceed- ings of the 12th ICLR
[12]

Albert Gu, Frederic Sala, Beliz Gunel, and Christopher Ré. 2019. Learning mixed- curvature representations in products of model spaces. In Proceedings of the 7th ICLR

2019
[13]

Aditya Grover and Jure Leskovec. 2016. node2vec: Scalable Feature Learning for Networks. In Proceedings of the 22nd SIGKDD . ACM, 855–864

2016
[14]

Zhenyu Hou, Yufei He, Yukuo Cen, Xiao Liu, Yuxiao Dong, Evgeny Kharlamov, and Jie Tang. 2023. GraphMAE2: A Decoding-Enhanced Masked Self-Supervised Graph Learner. In Proceedings of the ACM Web Conference (WWW) . 737–746

2023
[15]

Hamilton, Zhitao Ying, and Jure Leskovec

William L. Hamilton, Zhitao Ying, and Jure Leskovec. 2017. Inductive Represen- tation Learning on Large Graphs. In Advances in the 30th NeurIPS . 1024–1034

2017
[16]

Qian Huang, Hongyu Ren, Peng Chen, Gregor Krzmanc, Daniel Zeng, Percy Liang, and Jure Leskovec. 2023. PRODIGY: Enabling In-context Learning Over Graphs. In Advances in 36th NeurIPS

2023
[17]

Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta, and Jure Leskovec. 2020. Open Graph Benchmark: Datasets for Machine Learning on Graphs. In Advances in the 33rd NeurIPS , Vol. 33. 22118– 22133

2020
[18]

Ahmad Khajenezhad, Seyed Ali Osia, Mahmood Karimian, and Hamid Beigy
[19]

Wei Ju, Yifan Wang, Yifang Qin, Zhengyang Mao, Zhiping Xiao, Junyu Luo, Junwei Yang, Yiyang Gu, Dongjie Wang, Qingqing Long, Siyu Yi, Xiao Luo, and Ming Zhang. 2024. Towards Graph Contrastive Learning: A Survey and Beyond. CoRR abs/2405.11868 (2024). arXiv:2405.11868

2024 arXiv
[20]

Marc Law. 2021. Ultrahyperbolic Neural Networks. In Advances in the 34th NeurIPS. 22058–22069

2021
[21]

Hao Liu, Jiarui Feng, Lecheng Kong, Ningyue Liang, Dacheng Tao, Yixin Chen, and Muhan Zhang. 2024. One For All: Towards Training One Graph Model For All Classification Tasks. In Proceedings of the 12th ICLR

2024
[22]

Kipf and Max Welling

Thomas N. Kipf and Max Welling. 2017. Semi-Supervised Classification with Graph Convolutional Networks. In Proceedings of the 5th ICLR

2017
[23]

Haitao Mao, Zhikai Chen, Wenzhuo Tang, Jianan Zhao, Yao Ma, Tong Zhao, Neil Shah, Mikhail Galkin, and Jiliang Tang. 2024. Position: Graph Foundation Models Are Already Here. In Proceedings of the 41st ICML

2024
[24]

Kaplan, and Jiawei Han

Yu Meng, Jiaxin Huang, Guangyuan Wang, Chao Zhang, Honglei Zhuang, Lance M. Kaplan, and Jiawei Han. 2019. Spherical Text Embedding. In Advances in the 32nd NeurIPS . 8206–8215

2019
[25]

Qi Liu, Maximilian Nickel, and Douwe Kiela. 2019. Hyperbolic Graph Neural Networks. In Advances in the 32nd NeurIPS . 8228–8239

2019
[26]

Maximilian Nickel and Douwe Kiela. 2018. Learning Continuous Hierarchies in the Lorentz Model of Hyperbolic Geometry. In Proceedings of the 35th ICML , Vol. 80. PMLR, 3776–3785

2018
[27]

OpenAI. 2023. GPT-4 Technical Report. CoRR (2023)

2023
[28]

Péter Mernyei and Catalina Cangea. 2020. Wiki-CS: A Wikipedia-Based Bench- mark for Graph Neural Networks. CoRR (2020)

2020
[29]

Leonardo Filipe Rodrigues Ribeiro, Pedro H. P. Saverese, and Daniel R. Figueiredo
[30]

Benedek Rozemberczki, Carl Allen, and Rik Sarkar. 2021. Multi-Scale attributed node embedding. J. Complex Networks 9, 2 (2021)

2021
[31]

Peter Petersen. 2016. Riemannian Geometry, 3rd edition. Springer-Verlag

2016
[32]

Oleksandr Shchur, Maximilian Mumme, Aleksandar Bojchevski, and Stephan Günnemann. 2018. Pitfalls of Graph Neural Network Evaluation. CoRR (2018)

2018
[33]

Li Sun, Jingbin Hu, Suyang Zhou, Zhenhao Huang, Junda Ye, Hao Peng, Zhengtao Yu, and Philip S. Yu. 2024. RicciNet: Deep Clustering via A Riemannian Generative Model. In Proceedings of the ACM Web Conference (WWW) . 4071–4082

2024
[34]

Li Sun, Zhenhao Huang, Hao Peng, Yujie Wang, Chunyang Liu, and Philip S. Yu. 2024. LSEnet: Lorentz Structural Entropy Neural Network for Deep Graph Clustering. In Proceedings of the 41st ICML

2024
[35]

Rik Sarkar. 2011. Low Distortion Delaunay Embedding of Trees in Hyperbolic Plane. In Proceedings of the 19th International Symposium on Graph Drawing . Springer, 355–366

2011
[36]

Li Sun, Zhenhao Huang, Zixi Wang, Feiyang Wang, Hao Peng, and Philip S. Yu. 2024. Motif-aware Riemannian Graph Neural Network with Generative- Contrastive Learning. In Proceedings of the 38th AAAI . 9044–9052

2024
[37]

Li Sun, Zhenhao Huang, Hua Wu, Junda Ye, Hao Peng, Zhengtao Yu, and Philip S. Yu. 2023. DeepRicci: Self-supervised Graph Structure-Feature Co-Refinement for Alleviating Over-squashing. In Proceedings of the 23rd ICDM . 558–567

2023
[38]

Li Sun, Feiyang Wang, Junda Ye, Hao Peng, and Philip S. Yu. 2023. Congregate: Contrastive Graph Clustering in Curvature Spaces. In Proceedings of the 32nd IJCAI. 2296–2305

2023
[39]

Li Sun, Zhenhao Huang, Qiqi Wan, Hao Peng, and Philip S. Yu. 2024. Spiking Graph Neural Network on Riemannian Manifolds. In Advances in NeurIPS

2024
[40]

Li Sun, Junda Ye, Hao Peng, and Philip S. Yu. 2022. A Self-supervised Riemannian GNN with Time Varying Curvature for Temporal Graph Learning. InProceedings of the 31st CIKM . 1827–1836

2022
[41]

Li Sun, Ziheng Zhang, Zixi Wang, Yujie Wang, Qiqi Wan, Hao Li, Hao Peng, and Philip S. Yu. 2025. Pioneer: Physics-informed Riemannian Graph ODE for Entropy-increasing Dynamics. In Proceedings of the 39th AAAI

2025
[42]

Li Sun, Zhongbao Zhang, Junda Ye, Hao Peng, Jiawei Zhang, Sen Su, and Philip S. Yu. 2022. A Self-Supervised Mixed-Curvature Graph Neural Network. In Proceed- ings of the 36th AAAI . 4146–4155

2022
[43]

Li Sun, Junda Ye, Hao Peng, Feiyang Wang, and Philip S. Yu. 2023. Self-Supervised Continual Graph Learning in Adaptive Riemannian Spaces. In Proceedings of the 37th AAAI. 4633–4642

2023
[44]

Xiangguo Sun, Hong Cheng, Jia Li, Bo Liu, and Jihong Guan. 2023. All in One: Multi-Task Prompting for Graph Neural Networks. In Proceedings of the 29th SIGKDD. 2120–2131

2023
[45]

Petar Velickovic, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. 2018. Graph Attention Networks. In Proceedings of the 6th ICLR

2018
[46]

Hamilton, Pietro Liò, Yoshua Bengio, and R

Petar Velickovic, William Fedus, William L. Hamilton, Pietro Liò, Yoshua Bengio, and R. Devon Hjelm. 2019. Deep Graph Infomax. In Proceedings of the 7th ICLR

2019
[47]

Li Sun, Zhongbao Zhang, Jiawei Zhang, Feiyang Wang, Hao Peng, Sen Su, and Philip S. Yu. 2021. Hyperbolic Variational Graph Neural Network for Modeling Dynamic Graphs. In Proceedings of the 35th AAAI . 4375–4383

2021
[48]

Souza Jr., Tianyi Zhang, Christopher Fifty, Tao Yu, and Kilian Q

Felix Wu, Amauri H. Souza Jr., Tianyi Zhang, Christopher Fifty, Tao Yu, and Kilian Q. Weinberger. 2019. Simplifying Graph Convolutional Networks. In Proceedings of the 36th ICML , Vol. 12858. 35–43

2019
[49]

Lianghao Xia and Chao Huang. 2024. AnyGraph: Graph Foundation Model in the Wild. arXiv:2408.10700

2024 arXiv
[50]

Lianghao Xia, Ben Kao, and Chao Huang. 2024. OpenGraph: Towards Open Graph Foundation Models. In Proceedings of the EMNLP

2024
[51]

Yujie Wang, Shuo Zhang, Junda Ye, Hao Peng, and Li Sun. 2024. A Mixed- Curvature Graph Diffusion Model. In Proceedings of the 33rd CIKM . ACM, 2482– 2492

2024
[52]

Menglin Yang, Harshit Verma, Delvin Ce Zhang, Jiahong Liu, Irwin King, and Rex Ying. 2024. Hypformer: Exploring Efficient Transformer Fully in Hyperbolic Space. In Proceedings of the 30th SIGKDD . ACM, 3770–3781

2024
[53]

Cohen, and Ruslan Salakhutdinov

Zhilin Yang, William W. Cohen, and Ruslan Salakhutdinov. 2016. Revisiting Semi-Supervised Learning with Graph Embeddings. In Proceedings of the 33rd ICML. 40–48

2016
[54]

Junda Ye, Zhongbao Zhang, Li Sun, Yang Yan, Feiyang Wang, and Fuxin Ren
[55]

Bo Xiong, Shichao Zhu, Nico Potyka, Shirui Pan, Chuan Zhou, and Steffen Staab
[56]

In Advances in the 35th NeurIPS

Pseudo-Riemannian Graph Convolutional Networks. In Advances in the 35th NeurIPS
[57]

Yiding Zhang, Xiao Wang, Chuan Shi, Nian Liu, and Guojie Song. 2021. Lorentzian Graph Convolutional Networks. In Proceedings of the ACM Web Conference WWW ’25, April 28-May 2, 2025, Sydney, NSW, Australia Li Sun et. al. (WWW). 1249–1261

2021
[58]

Haihong Zhao, Aochuan Chen, Xiangguo Sun, Hong Cheng, and Jia Li. 2024. All in One and One for All: A Simple yet Effective Method towards Cross-domain Graph Pretraining. In Proceedings of the 30th SIGKDD . 4443–4454

2024
[59]

Jianan Zhao, Hesham Mostafa, Mikhail Galkin, Michael Bronstein, Zhaocheng Zhu, and Jian Tang. 2024. GraphAny: A Foundation Model for Node Classification on Any Graph. arXiv:2405.20445

2024 arXiv
[60]

Wenqiao Zhu, Yesheng Xu, Xin Huang, Qiyang Min, and Xun Zhou. 2022. Spher- ical Graph Embedding for Item Retrieval in Recommendation System. In Proceed- ings of the 31st CIKM . ACM, 4752–4756

2022
[61]

Tao Yu and Chris De Sa. 2023. Random Laplacian Features for Learning with Hyperbolic Space. In Proceedings of the 11th ICLR . OpenReview.net, 1–23

2023
[62]

Prasanna

Hanqing Zeng, Hongkuan Zhou, Ajitesh Srivastava, Rajgopal Kannan, and Vik- tor K. Prasanna. 2020. GraphSAINT: Graph Sampling Based Inductive Learning Method. In Proceedings of the 8th ICLR

2020
[67]

Yanqiao Zhu, Yichen Xu, Feng Yu, Qiang Liu, Shu Wu, and Liang Wang. 2021. Graph Contrastive Learning with Adaptive Augmentation. In Proceedings of the ACM Web Conference (WWW). ACM / IW3C2, 2069–2080. A Notation Table Table 5: Importation Notations. Notation Description M, 𝔤 A...

2021
[2017]

In Proceedings of the 23rd SIGKDD

struc2vec: Learning Node Representations from Structural Identity. In Proceedings of the 23rd SIGKDD . ACM, 385–394
[2022]

CoRR abs/2203.13655 (2022)

Gransformer: Transformer-based Graph Generation. CoRR abs/2203.13655 (2022). arXiv:2203.13655

2022 arXiv
[2023]

In Proceedings of the ACM Web Conference (WWW)

SINCERE: Sequential Interaction Networks representation learning on Co-Evolving RiEmannian manifolds. In Proceedings of the ACM Web Conference (WWW). 360–371
[2024]

In Proceedings of the 41st ICML

LLaGA: Large Language and Graph Assistant. In Proceedings of the 41st ICML

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.

[1] [1]

Gregor Bachmann, Gary Bécigneul, and Octavian Ganea. 2020. Constant Curva- ture Graph Convolutional Networks. In Proceedings of the 37th ICML , Vol. 119. PMLR, 486–496

work page 2020

[2] [2]

Dominique Beaini, Shenyang Huang, and Joao Alex Cunha et. al. 2024. Towards Foundational Models for Molecular Learning on Large-Scale Multi-Task Datasets. In Proceedings of the 12th ICLR . OpenReview.net

work page 2024

[3] [3]

Ines Chami, Zhitao Ying, Christopher Ré, and Jure Leskovec. 2019. Hyperbolic Graph Convolutional Neural Networks. In Advances in the 32nd NeurIPS . 4869– 4880

work page 2019

[4] [4]

Yupeng Chang, Xu Wang, and Jindong Wang. 2024. A Survey on Evaluation of Large Language Models. ACM Trans. Intell. Syst. Technol. 15, 3 (2024), 39:1–39:45

work page 2024

[5] [5]

Runjin Chen, Tong Zhao, Ajay Kumar Jaiswal, Neil Shah, and Zhangyang Wang

[6] [6]

Weize Chen, Xu Han, Yankai Lin, Hexu Zhao, Zhiyuan Liu, Peng Li, Maosong Sun, and Jie Zhou. 2022. Fully Hyperbolic Neural Networks. In Proceedings of the 60th ACL. ACL, 5672–5686

work page 2022

[7] [7]

Haitz Sáez de Ocáriz Borde, Anees Kazi, Federico Barbero, and Pietro Liò. 2023. Latent Graph Inference using Product Manifolds. In Proceedings of the 11th ICLR . OpenReview.net

work page 2023

[8] [8]

Michaël Defferrard, Xavier Bresson, and Pierre Vandergheynst. 2016. Convo- lutional Neural Networks on Graphs with Fast Localized Spectral Filtering. In Advances in the 29th NeurIPS . 3837–3845

work page 2016