Pith. sign in

REVIEW 3 major objections 5 minor 53 references

Train Small, Deploy Large: Zero-Shot GNN Transfer Through Geometric Renormalization

T0 review · 3 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

Pith's one-line read Training a GNN on a 32x-smaller geometrically renormalized replica of a graph transfers to the full graph with little accuracy loss.

desk verdict Solid empirical paper on GR-based zero-shot GNN transfer; the transfer result is real, but the structural-similarity interpretation is confounded by label-pure coarse blocks and needs a feature-only baseline. read the letter →

arxiv 2607.27767 v1 pith:WMIYVDQG submitted 2026-07-30 cs.LG cs.AIphysics.soc-ph

classification cs.LGcs.AIphysics.soc-ph
keywords graphneuralnetworksgeometricrenormalizationzero-shottransfercoarseninghyperbolicembeddingsscaleinvariancenodeclassificationnetworkgeometry
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 neural network can be trained on a heavily coarse-grained replica of a network and then deployed directly on the original full-scale network, without any retraining, while keeping most of the predictive accuracy. The coarse-graining is done by geometric renormalization, a scale transformation that preserves the local neighborhood statistics — degree distribution, clustering, and connection probabilities — that message-passing GNNs rely on. The authors test this on synthetic networks generated from a hyperbolic random graph model and on eight real-world networks, with three GNN architectures, and find that transfer accuracy stays high even at 32-fold compression, while random aggregation of nodes destroys transfer. They also show that models trained at different scales learn aligned representations and follow similar training trajectories, indicating that the effective learning problem is approximately scale-invariant. If the claim holds, structural similarity matters more than network size for GNN transferability, which would make large-graph learning much cheaper and open the way to scale-equivariant graph architectures.

What carries the argument

The central object is geometric renormalization (GR), a scale transformation that takes a hyperbolic embedding of the network, merges blocks of r consecutive nodes in the similarity space into supernodes with renormalized hidden degrees and angular positions, and rescales links according to the connection probability of the hidden-metric-space model, then prunes edges to match the average degree. Its role is to generate a hierarchy of coarse-grained graphs that preserve the degree distribution, clustering spectrum, and other local statistics that message-passing GNNs depend on, so that training on a small replica and testing on the original becomes a valid zero-shot transfer. The protocol al

What would settle it

Run the same zero-shot protocol but replace the mean-aggregated supernode features with a random constituent node's features (or a different pooling, e.g., median); if transfer accuracy drops sharply, the effect is carried by feature averaging rather than by GR's topology preservation, and the paper's stated limitation is load-bearing.

Watch

Extended reading notes

Core claim

The paper's central claim is that GNN weights trained on a geometrically renormalized coarse graph can be transferred zero-shot to the original graph with much of the full-scale performance retained. Across synthetic and real networks, for GCN, GraphSAGE, and GAT, transfer test accuracy remains close to the original-scale accuracy at renormalization levels up to ℓ=5 (a 32× compression), while training time drops by up to ~20×. The random baseline, which aggregates nodes without respecting the latent geometry, degrades substantially, showing that the transfer is due to the structure-preserving nature of GR rather than to smaller size. Additionally, centered kernel alignment and orthogonal Pro

Load-bearing premise

The result rests on the assumption that averaging node features and majority-voting labels within each GR block preserves the alignment between features, labels, and geometry that the GNN uses, because the paper renormalizes structure but not features.

Editorial extensions

If this is right

  • GNN training on large graphs can be made substantially cheaper: training on a 32× compressed replica and transferring weights yields most of the full-scale accuracy with up to ~20× shorter training time for GraphSAGE on synthetic networks.
  • Because random renormalization degrades transfer while GR does not, the result implies that preserving the structural statistics of the graph, not reducing its size, is what enables zero-shot weight transfer.
  • Representations learned at different GR scales are highly similar (high CKA and orthogonal Procrustes), and prediction trajectories stay aligned (low Jensen-Shannon divergence), indicating that the learned representation space and effective objective are approximately scale-invariant under GR.
  • In comparisons on a real network, GR achieves a smaller transfer-accuracy gap and lower combined topology-preservation error than Laplacian renormalization and edge-contraction pooling methods at comparable compression rates.
  • The accompanying GPU-accelerated implementation of the hyperbolic embedding step reduces its runtime by more than 400× on 10,000-node networks, making the protocol practical for networks that were previously too large to embed.

Reading between the lines

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

  • A natural testable extension: if the same zero-shot protocol works for link prediction and graph classification, the scale-invariance finding would generalize beyond node classification; the paper explicitly leaves these tasks open.
  • Because the paper renormalizes only topology, the protocol implicitly assumes features are approximately invariant under averaging; on graphs where features vary sharply within angular blocks, transfer may degrade—suggesting a feature renormalization scheme as the next algorithmic step.
  • The near-identical training trajectories under GR suggest that renormalization-based data augmentation—training on multiple GR scales simultaneously—could yield GNNs even less sensitive to scale shifts than any single-scale training.
  • If structural similarity truly outranks size for transferability, then the relevant complexity measure for GNN scaling is not node count but the fidelity with which a training graph reproduces the deployment graph's local geometry.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes a 'train-small, deploy-large' protocol in which a GNN is trained on a graph coarse-grained by geometric renormalization (GR) and the learned weights are transferred directly to the original graph without retraining. The authors evaluate this on synthetic HypBench graphs (SW and USW regimes) and eight real-world datasets using GCN, GraphSAGE, and GAT. They report that GR-preserved transfer accuracy is substantially higher than that of a random renormalization baseline, that representation-similarity measures (CKA, OP) and training-trajectory divergences (Jensen–Shannon) remain aligned across GR scales, and that GR outperforms LRG and two edge-pooling methods on PubMed. They also introduce cuMercator, a GPU implementation of Mercator. The central interpretive claim is that structural similarity, rather than network size, drives GNN transferability across scales.

Significance. If the central claim is established, the protocol offers a principled way to reduce training cost for large graphs and points toward scale-equivariant GNN architectures. The empirical scope is substantial: two synthetic regimes, eight real datasets, three architectures, ten random splits, and multiple baselines, including Random-T/Random-G partial ablations. The release of cuMercator is a useful methodological contribution. However, the key evidence for the structural-similarity interpretation is weakened by a confound between geometric block purity and supervised signal quality, as detailed below. The paper's contribution is therefore better viewed as an empirical demonstration that GR coarse-graining can support weight transfer, with the mechanistic explanation still open.

major comments (3)
  1. [Section 4.1 / 5.1 / Discussion] The GR-versus-Random comparison is confounded by feature/label purity. GR blocks are formed from angularly consecutive Mercator coordinates; on the homophilic datasets used here (HypBench α=6; Table 2 homophily 0.65–0.93), such blocks are predominantly single-class. Supernode features are block averages and labels are majority votes (Sec. 4.1), so the GR coarse training set is a systematically cleaner classification problem than the Random baseline, whose blocks mix classes. The claim in Sec. 6 that 'random aggregation substantially degrades transfer, demonstrating that preserving network structure is essential' does not follow, because the two protocols differ in the supervision signal as well as in topology. The partial baselines in App. A.8.3 (Random-T, Random-G) do not resolve this: both alter block purity together with coarse topology. To support the structural-similarity interpreta
  2. [Section 4.1 vs Section 7] The first stated limitation, 'we renormalize graph structure but not node features,' is inaccurate and obscures a design choice. The protocol does transform features: supernode features are the mean of constituent node features, and the GNN is trained on these averaged features but deployed on the original features. This is a feature-renormalization rule and is part of the transfer protocol, not an omission. The paper should treat the feature transform as a choice to be tested (e.g., compare mean vs. max vs. no aggregation) rather than as a limitation. This matters because the transfer gap may be driven by the feature transform rather than by the graph topology.
  3. [Abstract / Table 1 / Sec. 5.2] The blanket claim 'across synthetic and real-world networks, training on GR scaled-down replicas preserves much of the original-scale predictive performance' is not supported by the full real-world table. Table 1 shows WikiCS dropping from 0.83 to 0.64 at ℓ=5 (a 23% relative loss) and Cora from 0.87 to 0.71; Flickr stays at ~0.50 at all levels, i.e., no measurable transfer. While the text (Sec. 5.2) acknowledges that real-world transfer is weaker, the abstract and Discussion overstate the universality. Please qualify the claim to report the range of drops and the heterophilic failure case (Flickr, homophily 0.32).
minor comments (5)
  1. [Appendix A.2, Eq. (8)] There is a typographical error in the CKA denominator: '∥X ⊤ 0 X0|F ∥(X ↑ ℓ )⊤X ↑ ℓ ∥F' should read '∥X ⊤ 0 X0∥F ∥(X ↑ ℓ )⊤X ↑ ℓ ∥F'.
  2. [Section 5.2 / Appendix A.5] The comparison with LRG, MagEdgePool, and SpreadEdgePool is conducted on a single dataset (PubMed). The combined topology-preservation error in Eq. (13) normalizes each error by the maximum across methods, making the combined score dependent on the method set. The conclusion that GR provides the most faithful replica would be more persuasive with additional datasets and a fixed normalization convention.
  3. [Section 5.1 / Figure 3] Panels f and g use a GCN with r=32, while the main text says 'unless stated otherwise, we use r=2 in all experiments.' Specify that the trajectory panels use a different resolution, and clarify whether panels d/e use r=2 or r=32.
  4. [Abstract / Appendix A.6] The paper states 'We release cuMercator' but provides no URL or repository reference. Please add a link or note on availability; this is important for reproducibility.
  5. [Table 2] Flickr has homophily 0.32 and inferred β=1.01, yet it is included in the main real-world results despite the stated focus on homophilic graphs. This is not an error, but the paper should explicitly discuss Flickr as a heterophilic control case and temper the claim that GR preserves transfer on real networks.

Circularity Check

1 steps flagged · score 2.0 of 10

No significant circularity: the central GR-transfer claim is an empirical result checked against external benchmarks and random baselines; self-citation is present but not load-bearing, and the label-purity confound is a correctness risk, not a construction.

  1. self citation load bearing [Section 3 (Network geometry and geometric renormalization), applied throughout Section 4.1]
    "Geometric renormalization (GR) [14] is a technique that merges nearby nodes along the similarity subspace, producing a hierarchy of graph representations at progressively lower resolutions that preserve the multiscale organization of the original network."

    The core tool whose transfer-preserving properties are the paper's subject is introduced via [14] (García-Pérez, Boguñá, Serrano — co-authors here), the embeddings use Mercator [33,34] from the same group, and the synthetic benchmark HypBench [16] is also authored by Jankowski/Serrano/Boguñá. So the framework is substantially self-referential. The load, however, is not carried by the citations alone: the paper re-states GR's equations (Eqs. 5-6, App. A.1) and independently re-validates the preservation hypotheses in its own Figure 2b and Appendix A.8.4, while the transfer claim is judged on standard external benchmarks (Cora, PubMed, Computers, Photo, CS, Physics, WikiCS, Flickr) against GR-vs-Random contrasts. This self-citation is thus non-load-bearing, though worth a minor deduction.

full rationale

The central claim — a GNN trained on a GR-coarse-grained graph Gℓ transfers zero-shot to the original graph G0 — is not built into the construction. GR is a topology-only transformation (block merging plus metric pruning, Eqs. 5-6 and App. A.1) that never uses GNN weights, never uses transfer accuracy as an objective, and is not defined in terms of the GNN's success. Transfer test accuracy is measured on original test nodes after training only on Gℓ, so it is a genuine empirical outcome that could have failed; it does fail for the Random, Random-T, and Random-G baselines. No equation in the paper equals the transfer result to its inputs, and no fitted parameter is renamed as a prediction. GNN training is standard (GCN/GraphSAGE/GAT) on fixed, external benchmarks. Self-citation is dense (GR [14], Mercator [33,34], HypBench [16], disentanglement strategy [25]) but non-load-bearing because the paper independently validates GR's invariance properties and evaluates on external real networks. The strongest skeptical attack is the label-purity confound: GR blocks are angularly contiguous on homophilic data, so supernode majority-vote labels and averaged features (Sec. 4.1) make GR's coarse training problem cleaner than Random's; the paper itself admits in Sec. 5.1 that Random models 'continue to leverage the original node features,' and in Section 7 concedes 'we renormalize graph structure but not node features.' That is an alternative explanation and a correctness risk, not a circular derivation — there is no exhibited step where the GR-vs-Random accuracy gap reduces by construction to block purity. Similarly, the CKA/OP lifting in Sec. 4.2 assigns identical lifted vectors within each GR block, which can inflate representational similarity, but the same lifting is applied to both protocols and does not determine the central accuracy claim. Verdict: no significant circularity; score reflects only the self-referential provenance of the core tooling.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

No new particles, forces, or mathematical objects are introduced. The paper's assumptions are domain assumptions inherited from the S1/H2/Mercator line of work, plus the ad hoc pooling choices for features and labels. Free parameters are mostly inherited from the GR method and the embedding step, plus r.

free parameters (3)
  • renormalization resolution r = r=2 per level (main experiments); r=32 for the trajectory panels; other r values used to match compression in the method
    The coarsening strength is a user-chosen hyperparameter. The main results are shown as a function of level with fixed r=2, but the strong trajectory claim (Fig. 3f/g) uses r=32, so the most visually striking alignment is achieved at a different compression than the accuracy curves.
  • Mercator inferred parameters (beta, mu, hidden degrees, angular coordinates) = Inferred per real network via Mercator; includes model parameters beta and mu.
    For real networks, the entire GR flow depends on the hyperbolic embedding obtained by maximum-likelihood inference (O(N^2), approximated with negative sampling). If the embedding is inaccurate, the merging blocks are not the true geometric blocks. This is a fitted input, not a free parameter tuned to the transfer result, but it is load-bearing.
  • LRG temporal resolution parameter tau = tau in {0.5, 1.6, 2} for the comparison in Figure 5; tau=1.2 for Figure 6.
    Tuned by hand to match target compression levels in the method comparison; affects the comparison, not the central GR claim.
assumptions (4)
  • domain assumption The S1/H2 hidden-metric-space model is the correct generative description of the benchmark and real networks (Section 3, Eq. 1-2).
    GR's guarantee that coarse-graining preserves connection probabilities follows from the renormalizability of S1/H2. Its use on real networks assumes they admit a good Mercator embedding; the paper explicitly conditions on this ('as long as they admit a good embedding').
  • domain assumption Standard message-passing GNNs (GCN, GraphSAGE, GAT) at two layers with 32 hidden units suffice to probe transferability.
    Results are only shown for these architectures and sizes. No deep models or other GNN families are tested, so the scope of the claim is bounded by this choice.
  • domain assumption Preserving degree distribution, clustering spectrum, knn, average degree, and spectral gap is sufficient to preserve GNN transfer behavior.
    The paper validates GR topologically with these quantities and then interprets preserved transfer accuracy as caused by that preservation; this is correlational, not causal.
  • ad hoc to paper Majority-vote labels and averaged features on supernodes are a valid supervision and feature signal (Section 4.1).
    Chosen for convenience; the paper itself flags feature averaging as unprincipled (Section 7). No alternative pooling rules are tested.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Train Small, Deploy Large: Zero-Shot GNN Transfer Through Geometric Renormalization." pith.science (2026). https://pith.science/paper/WMIYVDQG

@misc{pith2026260727767,
  author       = {Pith},
  title        = {Pith review of: Train Small, Deploy Large: Zero-Shot GNN Transfer Through Geometric Renormalization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WMIYVDQG}},
  note         = {Machine review of arXiv:2607.27767}
}
read the original abstract

Graph neural networks (GNNs) can operate on large graphs but become infrastructure-sensitive at the scale of millions of nodes and typically require scalable training techniques for even larger graphs. This raises a central question: when can a model trained on a smaller, scaled-down replica of a graph be deployed on the full-resolution graph without retraining? We introduce a zero-shot transfer protocol in which a GNN is trained on a graph coarse-grained by geometric renormalization (GR), and the resulting weights are transferred directly to the original network. Across synthetic and real-world networks, training on GR scaled-down replicas preserves much of the original-scale predictive performance while significantly reducing training cost. We further find that learned representations and predictive trajectories remain aligned across scales. These findings suggest that structural similarity may be more important than network size in determining GNN transferability, opening a path toward scale-equivariant graph architectures.

Figures

Figures reproduced from arXiv: 2607.27767 by the authors.

Figure 1
Figure 1. Overview of the train-small, deploy-large protocol [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. (a) Schematic of geometric renormalization. Each layer is obtained by applying a renormalization step with resolution r, starting from the original network at l = 0. Orange nodes are placed on the circle, with sizes proportional to the logarithm of their hidden degrees, and solid lines denote links within each layer. Blue shaded regions indicate coarse-graining blocks, and dashed lines connect nodes to their superno… view at source ↗
Figure 3
Figure 3. Zero-shot GNN transfer on synthetic networks [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (19 more)
Figure 4
Figure 4. Figure 4: Zero-shot GNN transfer on the Photo dataset [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Combined topology preservation er￾ror as a function of the difference between the transfer and original test accuracies for differ￾ent coarse-graining methods for the PubMed dataset. Marker color denotes the coarse￾graining method, while marker size represents the comp…
Figure 6
Figure 6. Figure 6: Comparison of coarse-graining methods on the PubMed dataset. We report: (a) the complementary cumulative degree distribution, (b) the clustering spectrum, (c) the average-neighbor￾degree spectrum, (d) the average degree, (e) the average clustering coefficient, (f) the …
Figure 7
Figure 7. Figure 7: Computational performance of Mercator. (a) Embedding time as a function of network size for the CPU and GPU implementations. (b) Embedding time on the GPU using either exact maximum-likelihood (ML) calculations or negative sampling with S = 64. (c) c-score between the …
Figure 8
Figure 8. Figure 8: shows the topological validation of geometric renormalization in the ultra-small-world (USW) regime. Similarly to the small-world regime, the topological properties are preserved [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]
Figure 9
Figure 9. Figure 9: summarizes the zero-shot transfer results. 100 101 102 103 k 10−5 10−4 10−3 10−2 10−1 100 Pc(k) l = 0 l = 1 l = 2 l = 3 l = 4 101 102 103 k 0.0 0.2 0.4 0.6 0.8 c¯(k) 100 101 102 103 k 0 50 100 150 ¯knn(k) 0 20 ¯k 0 1 H 0 1 2 3 4 Renormalization level ` 0 1 µ L 2 [PITH…
Figure 10
Figure 10. Figure 10: Zero-shot GNN transfer on PubMed dataset [PITH_FULL_IMAGE:figures/full_fig_p018_10.png]
Figure 11
Figure 11. Figure 11: Zero-shot GNN transfer on CS dataset. See the caption of [PITH_FULL_IMAGE:figures/full_fig_p018_11.png]
Figure 12
Figure 12. Figure 12: Zero-shot GNN transfer on Physics dataset [PITH_FULL_IMAGE:figures/full_fig_p019_12.png]
Figure 13
Figure 13. Figure 13: Zero-shot GNN transfer on Computers dataset [PITH_FULL_IMAGE:figures/full_fig_p019_13.png]
Figure 14
Figure 14. Figure 14: Zero-shot GNN transfer on WikiCS dataset [PITH_FULL_IMAGE:figures/full_fig_p020_14.png]
Figure 15
Figure 15. Figure 15: Zero-shot GNN transfer on Flickr dataset [PITH_FULL_IMAGE:figures/full_fig_p020_15.png]
Figure 16
Figure 16. Figure 16: Zero-shot GNN transfer on Cora dataset. See the caption of [PITH_FULL_IMAGE:figures/full_fig_p021_16.png]
Figure 17
Figure 17. Figure 17: Zero-shot GNN transfer accuracy for SW and USW regimes, against Random-T [PITH_FULL_IMAGE:figures/full_fig_p021_17.png]
Figure 18
Figure 18. Figure 18: Zero-shot GNN transfer accuracy across all real-world datasets, against Random-T [PITH_FULL_IMAGE:figures/full_fig_p022_18.png]
Figure 19
Figure 19. Figure 19: Topological validation of GR for the real-world datasets considered (part 1 of 2). 23 [PITH_FULL_IMAGE:figures/full_fig_p023_19.png]
Figure 20
Figure 20. Figure 20: Topological validation of GR for the real-world datasets considered (part 2 of 2). 24 [PITH_FULL_IMAGE:figures/full_fig_p024_20.png]
Figure 21
Figure 21. Figure 21: Topological validation of the SW regime under three random variants of the renormaliza￾tion method. 25 [PITH_FULL_IMAGE:figures/full_fig_p025_21.png]
Figure 22
Figure 22. Figure 22: Topological validation of the SW regime under GR and three random variants-Random, Random-G, and Random-T of the renormalization method, shown for renormalization levels l = 1, 2, 3, 4. Dashed black lines indicate the reference values of the original (non-renormalized…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

53 extracted references · 1 canonical work pages

  1. [1]

    Kipf and Max Welling

    Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. InInternational Conference on Learning Representations, 2017. URL https: //openreview.net/forum?id=SJU4ayYgl. 1, 5

  2. [2]

    Inductive representation learning on large graphs

    Will Hamilton, Zhitao Ying, and Jure Leskovec. Inductive representation learning on large graphs. In I. Guyon, U. V on Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett, editors,Advances in Neural Information Processing Systems, volume 30. Curran Associates, Inc., 2017. URL https://proceedings.neurips.cc/paper_files/paper/ 2017/fi...

  3. [3]

    From local structures to size generalization in graph neural networks

    Gilad Yehudai, Ethan Fetaya, Eli Meirom, Gal Chechik, and Haggai Maron. From local structures to size generalization in graph neural networks. InInternational Conference on Machine Learning, pages 11975–11986. PMLR, 2021. 1, 3

  4. [4]

    Graphon neural networks and the trans- ferability of graph neural networks.Advances in Neural Information Processing Systems, 33: 1702–1712, 2020

    Luana Ruiz, Luiz Chamon, and Alejandro Ribeiro. Graphon neural networks and the trans- ferability of graph neural networks.Advances in Neural Information Processing Systems, 33: 1702–1712, 2020. 2, 3

  5. [5]

    The structure and function of complex networks.SIAM Review, 45(2): 167–256, 2003

    Mark EJ Newman. The structure and function of complex networks.SIAM Review, 45(2): 167–256, 2003. 2

  6. [6]

    Graph neural networks are not continuous across graph resolutions.arXiv preprint arXiv:2605.31315, 2026

    Christian Koke, Yuesong Shen, Abhishek Saroha, Marvin Eisenberger, Bastian Rieck, Michael Bronstein, and Daniel Cremers. Graph neural networks are not continuous across graph resolutions.arXiv preprint arXiv:2605.31315, 2026. 2, 3, 9

  7. [7]

    Graph reduction with spectral and cut guarantees.Journal of Machine Learning Research, 20(116):1–42, 2019

    Andreas Loukas. Graph reduction with spectral and cut guarantees.Journal of Machine Learning Research, 20(116):1–42, 2019. 2

  8. [8]

    Graph coarsening: from scientific computing to machine learning.SeMA Journal, 79(1):187–223, 2022

    Jie Chen, Yousef Saad, and Zechen Zhang. Graph coarsening: from scientific computing to machine learning.SeMA Journal, 79(1):187–223, 2022. 2

Show all 53 references
  1. [9]

    Graph coarsening with message-passing guarantees.Ad- vances in Neural Information Processing Systems, 37:114902–114927, 2024

    Antonin Joly and Nicolas Keriven. Graph coarsening with message-passing guarantees.Ad- vances in Neural Information Processing Systems, 37:114902–114927, 2024. 2, 3

  2. [10]

    Self-similarity of complex networks and hidden metric spaces.Physical Review Letters, 100(7):078701, 2008

    M Ángeles Serrano, Dmitri Krioukov, and Marián Boguná. Self-similarity of complex networks and hidden metric spaces.Physical Review Letters, 100(7):078701, 2008. 2, 3

  3. [11]

    Network geometry.Nature Reviews Physics, 3(2):114–135, 2021

    Marian Boguna, Ivan Bonamassa, Manlio De Domenico, Shlomo Havlin, Dmitri Krioukov, and M Ángeles Serrano. Network geometry.Nature Reviews Physics, 3(2):114–135, 2021. 2

  4. [12]

    Hyperbolic geometry of complex networks.Physical Review E—Statistical, Nonlinear , and Soft Matter Physics, 82(3):036106, 2010

    Dmitri Krioukov, Fragkiskos Papadopoulos, Maksim Kitsak, Amin Vahdat, and Marián Boguná. Hyperbolic geometry of complex networks.Physical Review E—Statistical, Nonlinear , and Soft Matter Physics, 82(3):036106, 2010. 2, 3

  5. [13]

    Cambridge University Press, 2022

    M Ángeles Serrano and Marián Boguñá.The Shortest Path to Network Geometry: A Practical Guide to Basic Models and Applications. Cambridge University Press, 2022. 2, 3

  6. [14]

    Multiscale unfolding of real networks by geometric renormalization.Nature Physics, 14(6):583–589, 2018

    Guillermo García-Pérez, Marián Boguñá, and M Ángeles Serrano. Multiscale unfolding of real networks by geometric renormalization.Nature Physics, 14(6):583–589, 2018. 2, 3, 4, 13, 15

  7. [15]

    Similarity of neural network representations revisited

    Simon Kornblith, Mohammad Norouzi, Honglak Lee, and Geoffrey Hinton. Similarity of neural network representations revisited. InInternational conference on machine learning, pages 3519–3529. PMLR, 2019. 2, 5

  8. [16]

    Ángeles Serrano, and Marián Boguñá

    Roya Aliakbarisani, Robert Jankowski, M. Ángeles Serrano, and Marián Boguñá. Hypbench: Hyperbolic benchmark for graph neural network performance.IEEE Transactions on Neural Networks and Learning Systems, pages 1–14, 2026. doi: 10.1109/TNNLS.2026.3697597. 2, 4, 5

  9. [17]

    Hierarchical graph representation learning with differentiable pooling.Advances in neural information processing systems, 31, 2018

    Zhitao Ying, Jiaxuan You, Christopher Morris, Xiang Ren, Will Hamilton, and Jure Leskovec. Hierarchical graph representation learning with differentiable pooling.Advances in neural information processing systems, 31, 2018. 2

  10. [18]

    Graph u-nets

    Hongyang Gao and Shuiwang Ji. Graph u-nets. In Kamalika Chaudhuri and Ruslan Salakhutdi- nov, editors,Proceedings of the 36th International Conference on Machine Learning, volume 97 ofProceedings of Machine Learning Research, pages 2083–2092. PMLR, 09–15 Jun 2019. URLhttps://p...

  11. [19]

    Understanding pooling in graph neural networks.IEEE transactions on neural networks and learning systems, 35(2):2708–2718, 2022

    Daniele Grattarola, Daniele Zambon, Filippo Maria Bianchi, and Cesare Alippi. Understanding pooling in graph neural networks.IEEE transactions on neural networks and learning systems, 35(2):2708–2718, 2022. 2

  12. [20]

    Geometry-aware edge pooling for graph neural networks.Advances in Neural Information Processing Systems, 38: 157770–157808, 2026

    Katharina Limbeck, Lydia Mezrag, Guy Wolf, and Bastian Rieck. Geometry-aware edge pooling for graph neural networks.Advances in Neural Information Processing Systems, 38: 157770–157808, 2026. 8, 14

  13. [21]

    Graph pooling via ricci flow.Transactions on Machine Learning Research, 2024

    Amy Feng and Melanie Weber. Graph pooling via ricci flow.Transactions on Machine Learning Research, 2024. 2

  14. [22]

    Scaling up graph neural networks via graph coarsening

    Zengfeng Huang, Shengzhong Zhang, Chong Xi, Tang Liu, and Min Zhou. Scaling up graph neural networks via graph coarsening. InProceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD ’21), pages 675–684, 2021. doi: 10.1145/ 3447548.3467256. 2

  15. [23]

    Network renormalization.Nature Reviews Physics, 7(4):203–219, 2025

    Andrea Gabrielli, Diego Garlaschelli, Subodh P Patil, and M Ángeles Serrano. Network renormalization.Nature Reviews Physics, 7(4):203–219, 2025. 3

  16. [24]

    Self-similarity of complex networks

    Chaoming Song, Shlomo Havlin, and Hernan A Makse. Self-similarity of complex networks. Nature, 433(7024):392–395, 2005. 3

  17. [25]

    Geometric renormalization unravels self-similarity of the multiscale human connectome

    Muhua Zheng, Antoine Allard, Patric Hagmann, Yasser Alemán-Gómez, and M Ángeles Ser- rano. Geometric renormalization unravels self-similarity of the multiscale human connectome. Proceedings of the National Academy of Sciences, 117(33):20244–20253, 2020. 3, 5

  18. [26]

    Geometric renormalization of weighted networks.Communications Physics, 7(1):97, 2024

    Muhua Zheng, Guillermo García-Pérez, Marián Boguñá, and M Ángeles Serrano. Geometric renormalization of weighted networks.Communications Physics, 7(1):97, 2024. 3

  19. [27]

    Laplacian renormaliza- tion group for heterogeneous networks.Nature Physics, 19(3):445–450, 2023

    Pablo Villegas, Tommaso Gili, Guido Caldarelli, and Andrea Gabrielli. Laplacian renormaliza- tion group for heterogeneous networks.Nature Physics, 19(3):445–450, 2023. 3, 8, 14

  20. [28]

    Higher-order laplacian renormalization.Nature Physics, 21(4):661–668, 2025

    Marco Nurisso, Marta Morandini, Maxime Lucas, Francesco Vaccarino, Tommaso Gili, and Giovanni Petri. Higher-order laplacian renormalization.Nature Physics, 21(4):661–668, 2025. 3

  21. [29]

    Multiscale network renormalization: Scale-invariance without geometry.Physical Review Research, 5(4):043101, 2023

    Elena Garuccio, Margherita Lalli, and Diego Garlaschelli. Multiscale network renormalization: Scale-invariance without geometry.Physical Review Research, 5(4):043101, 2023. 3

  22. [30]

    A metadata-driven approach to understand graph neural networks.Advances in Neural Information Processing Systems, 36:15320–15340, 2023

    Ting Wei Li, Qiaozhu Mei, and Jiaqi Ma. A metadata-driven approach to understand graph neural networks.Advances in Neural Information Processing Systems, 36:15320–15340, 2023. 3

  23. [31]

    Performance heterogeneity in graph neural networks: Lessons for architecture design and preprocessing.arXiv preprint arXiv:2503.00547, 2025

    Lukas Fesser and Melanie Weber. Performance heterogeneity in graph neural networks: Lessons for architecture design and preprocessing.arXiv preprint arXiv:2503.00547, 2025. 3

  24. [32]

    Transferability properties of graph neural networks.IEEE Transactions on Signal Processing, 71:3474–3489, 2023

    Luana Ruiz, Luiz FO Chamon, and Alejandro Ribeiro. Transferability properties of graph neural networks.IEEE Transactions on Signal Processing, 71:3474–3489, 2023. 3

  25. [33]

    Mercator: uncovering faithful hyperbolic embeddings of complex networks.New Journal of Physics, 21 (12):123033, 2019

    Guillermo García-Pérez, Antoine Allard, M Ángeles Serrano, and Marián Boguñá. Mercator: uncovering faithful hyperbolic embeddings of complex networks.New Journal of Physics, 21 (12):123033, 2019. 3, 5, 16

  26. [34]

    The d-mercator method for the multidimensional hyperbolic embedding of real networks.Nature Communica- tions, 14(1):7585, 2023

    Robert Jankowski, Antoine Allard, Marian Boguna, and M Ángeles Serrano. The d-mercator method for the multidimensional hyperbolic embedding of real networks.Nature Communica- tions, 14(1):7585, 2023. 3, 5, 9, 16

  27. [35]

    Ángeles Serrano

    Marian Boguñá, Dmitri Krioukov, Pedro Almagro, and M. Ángeles Serrano. Small worlds and clustering in spatial networks.Physical Review Research, 2(2):023040, Apr 2020. doi: 10.1103/PhysRevResearch.2.023040. 5, 14

  28. [36]

    Collective classification in network data.AI Magazine, 29(3):93–106, 2008

    Prithviraj Sen, Galileo Namata, Mustafa Bilgic, Lise Getoor, Brian Galligher, and Tina Eliassi- Rad. Collective classification in network data.AI Magazine, 29(3):93–106, 2008. 5

  29. [37]

    Revisiting semi-supervised learning with graph embeddings

    Zhilin Yang, William Cohen, and Ruslan Salakhudinov. Revisiting semi-supervised learning with graph embeddings. InInternational conference on machine learning, pages 40–48. PMLR,

  30. [38]

    Pitfalls of graph neural network evaluation.arXiv preprint arXiv:1811.05868, 2018

    Oleksandr Shchur, Max Mumme, Aleksandar Bojchevski, and Stephan Günnemann. Pitfalls of graph neural network evaluation.arXiv preprint arXiv:1811.05868, 2018. doi: 10.48550/arXiv. 1811.05868. 5 11 Train Small, Deploy Large: Zero-Shot GNN Transfer Through Geometric Renormalization

  31. [39]

    Wiki-cs: A wikipedia-based benchmark for graph neural networks.arXiv preprint arXiv:2007.02901, 2020

    Péter Mernyei and C˘at˘alina Cangea. Wiki-cs: A wikipedia-based benchmark for graph neural networks.arXiv preprint arXiv:2007.02901, 2020. 5

  32. [40]

    GraphSAINT: Graph sampling based inductive learning method

    Hanqing Zeng, Hongkuan Zhou, Ajitesh Srivastava, Rajgopal Kannan, and Viktor Prasanna. GraphSAINT: Graph sampling based inductive learning method. InInternational Conference on Learning Representations (ICLR), 2020. 5

  33. [41]

    Scale-free networks well done.Physical Review Research, 1(3):033034, 2019

    Ivan V oitalov, Pim Van Der Hoorn, Remco Van Der Hofstad, and Dmitri Krioukov. Scale-free networks well done.Physical Review Research, 1(3):033034, 2019. 5

  34. [42]

    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,

  35. [43]

    Matthias Fey and Jan E. Lenssen. Fast graph representation learning with PyTorch Geometric. InICLR Workshop on Representation Learning on Graphs and Manifolds, 2019. 5

  36. [44]

    Algorithms for learning kernels based on centered alignment.The Journal of Machine Learning Research, 13:795–828, 2012

    Corinna Cortes, Mehryar Mohri, and Afshin Rostamizadeh. Algorithms for learning kernels based on centered alignment.The Journal of Machine Learning Research, 13:795–828, 2012. 5

  37. [45]

    On kernel-target alignment.Advances in neural information processing systems, 14, 2001

    Nello Cristianini, John Shawe-Taylor, Andre Elisseeff, and Jaz Kandola. On kernel-target alignment.Advances in neural information processing systems, 14, 2001. 5

  38. [46]

    Grounding representation similarity through statistical testing.Advances in Neural Information Processing Systems, 34:1556–1568,

    Frances Ding, Jean-Stanislas Denain, and Jacob Steinhardt. Grounding representation similarity through statistical testing.Advances in Neural Information Processing Systems, 34:1556–1568,

  39. [47]

    Generalized shape metrics on neural representations.Advances in neural information processing systems, 34:4738–4750,

    Alex H Williams, Erin Kunz, Simon Kornblith, and Scott Linderman. Generalized shape metrics on neural representations.Advances in neural information processing systems, 34:4738–4750,

  40. [48]

    Spectral gap of random hyperbolic graphs and related param- eters.The Annals of Applied Probability, 28(2):941 – 989, 2018

    Marcos Kiwi and Dieter Mitsche. Spectral gap of random hyperbolic graphs and related param- eters.The Annals of Applied Probability, 28(2):941 – 989, 2018. doi: 10.1214/17-AAP1323. URLhttps://doi.org/10.1214/17-AAP1323. 6

  41. [49]

    Feature-aware ultra-low dimensional reduction of real networks.npj Complexity, 1(1):13, 2024

    Robert Jankowski, Pegah Hozhabrierdi, Marián Boguñá, and M Ángeles Serrano. Feature-aware ultra-low dimensional reduction of real networks.npj Complexity, 1(1):13, 2024. 8

  42. [50]

    Án- geles Serrano

    Aina Ferrà Marcús, Robert Jankowski, Meritxell Vila-Miñana, Carles Casacuberta, and M. Án- geles Serrano. Chordless cycle filtrations for dimensionality detection in complex networks via topological data analysis.Nature Communications, 17(1):6105, 2026. 9

  43. [51]

    Detecting the ultra low dimensionality of real networks.Nature Communications, 13(1):6096, 2022

    Pedro Almagro, Marián Boguñá, and M Ángeles Serrano. Detecting the ultra low dimensionality of real networks.Nature Communications, 13(1):6096, 2022. 9

  44. [52]

    Machine learning meets complex networks via coalescent embedding in the hyperbolic space.Nature Communications, 8(1):1615, 2017

    Alessandro Muscoloni, Josephine Maria Thomas, Sara Ciucci, Ginestra Bianconi, and Carlo Vit- torio Cannistraci. Machine learning meets complex networks via coalescent embedding in the hyperbolic space.Nature Communications, 8(1):1615, 2017. 16 12 Train Small, Deploy Large: Zer...

  45. [2018]

    URLhttps://openreview.net/forum?id=rJXMpikCZ. 5

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.