pith. sign in

arxiv: 2605.18927 · v1 · pith:A5X4D65Jnew · submitted 2026-05-18 · 📊 stat.ML · cs.LG· math.PR

Bayesian Latent Space Models for Graphs Are Misspecified: Toward Robust Inference via Generalized Posteriors

Aldric Labarthe (CB , UNIGE) This is my paper

Pith reviewed 2026-05-20 08:34 UTC · model grok-4.3

classification 📊 stat.ML cs.LGmath.PR
keywords Bayesian inferencelatent space modelsgraph misspecificationgeneralized posteriorsnetwork embeddingslink predictiongeometric graphsmodel calibration
0
0 comments X

The pith

Bayesian latent space models for networks become overconfident when geometry assumptions fail.

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

The paper shows that Bayesian latent space models for graphs assume a specific geometry and link function that real networks frequently violate through geometric mismatch and structural anomalies. This places the true data-generating process outside the model class, so standard Bayesian inference produces overconfident and poorly calibrated results. To correct this, the authors introduce a generalized posterior framework together with Link-Sequential R-SafeBayes, which uses the conditional independence of network edges to estimate prequential risk and adjust the amount of posterior regularization on the fly. Tests on synthetic graphs with known mismatches and on real networks demonstrate better calibration, stronger link prediction, and a practical way to pick the most suitable latent geometry among Euclidean, spherical, and hyperbolic options. Readers would care because network data appear in many domains where reliable uncertainty estimates matter for downstream decisions.

Core claim

Bayesian latent space models for graphs rely on correct specification of both geometry and link function, but real-world networks often violate these assumptions by exhibiting geometric mismatch and structural anomalies that break standard metric properties. Such misspecification pushes the data-generating distribution outside the model class, causing Bayesian inference to become overconfident and poorly calibrated. The authors therefore propose a generalized posterior framework for random geometric graphs and introduce Link-Sequential R-SafeBayes, a method that exploits dyadic conditional independence to estimate prequential risk and adaptively tune posterior regularization.

What carries the argument

Link-Sequential R-SafeBayes, which exploits dyadic conditional independence to estimate prequential risk and thereby adaptively tunes the regularization strength of the generalized posterior.

If this is right

  • Standard Bayesian posteriors in latent space models produce overconfident uncertainty estimates on networks that deviate from the assumed geometry.
  • Generalized posteriors with adaptive regularization improve calibration and link-prediction accuracy on both synthetic and real networks.
  • The method supplies a data-driven criterion for choosing among Euclidean, spherical, and hyperbolic latent geometries.
  • Real networks can be represented more reliably once the posterior is allowed to account for model misspecification.

Where Pith is reading between the lines

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

  • The same risk-estimation idea could extend to other Bayesian models of relational data where conditional independence of observations holds.
  • Applying the procedure to temporal or multilayer networks could test whether adaptive regularization also improves forecasts in dynamic settings.
  • Practitioners may use the geometry-selection criterion to diagnose which embedding space best matches the dominant structure of a given empirical network.

Load-bearing premise

The approach assumes that dyadic conditional independence is strong enough to yield a reliable estimate of prequential risk that correctly guides how much regularization to apply.

What would settle it

Generate networks from a geometry different from the one used in fitting, apply standard Bayesian inference, and check whether the resulting posterior predictive intervals still achieve nominal coverage; failure to observe overconfidence would refute the central claim.

Figures

Figures reproduced from arXiv: 2605.18927 by Aldric Labarthe (CB, UNIGE).

Figure 1
Figure 1. Figure 1: Conceptual framework of geometric misspecification in Bayesian RGGs. (1 and 2) An observed network G and its Latent positions X in space U inferred via the likelihood p(G|X) and the posterior p(X|G). (3) Theoretical illustration of model misspecification: the model set M (e.g., Euclidean RGGs) is inherently non-convex. When the ground truth P ∗ (e.g., hyperbolic) lies outside M, the distribution P˜ that mi… view at source ↗
Figure 2
Figure 2. Figure 2: Predictive Performance vs. Likelihood Weighting η. Relative cumulative log-loss (left) and square-loss (right) across networks, normalized to the standard Bayesian baseline (η = 1.0). 0.1 0.2 0.3 0.4 0.5 Learning Rate ( ) 0.5 1.0 1.5 2.0 2.5 3.0 Prequential Log Loss Varying Number of Blocks (K) Optimal * =0.3 K=3 K=5 K=7 K=10 0.1 0.2 0.3 0.4 0.5 Learning Rate ( ) 1.0 1.1 1.2 1.3 1.4 1.5 Prequential Log Los… view at source ↗
Figure 3
Figure 3. Figure 3: Robustness of Link-Sequential R-SafeBayes to prequential hyperparameters and data ordering (Dolphins network). (Left) Varying the number of evaluation blocks K. (Middle) Adjusting the initial training fraction ptrain. (Right) Individual empirical trajectories across six randomized dyadic shufflings (grey) and their mean (blue) with ±1 standard deviation. 4.2 Mitigating Overconfidence and Improving Predicti… view at source ↗
Figure 4
Figure 4. Figure 4: SafeBayes Overfitting correction evaluated in the Euclidean case. The shaded area represents the overfitting gap where Standard Bayes (η = 1.0) minimizes in-sample error at the cost of sub-optimal out-of-sample prediction loss. SafeBayes (η < 1.0) identifies the global minimum, achieving an average 15.6% reduction in log. loss and 11.6% in sq. loss. 0.0-0.2 0.2-0.4 0.4-0.6 0.6-0.8 0.8-1.0 Std. Bayes Out-of… view at source ↗
Figure 5
Figure 5. Figure 5: Distribution of Predictive Improvement. SafeBayes mitigates catastrophic “Sure-but￾Wrong” predictions inherent in Euclidean latent space models. While a minor regularization tax is observed in low-error dyads, the cumulative information gain (blue line) is driven by massive penalty reductions in the high-error regime (Error > 0.8). 4.3 Robust Geometric Selection via Prequential Risk Beyond guarding against… view at source ↗
read the original abstract

Bayesian latent space models offer a principled approach to network representation, but rely on correct specification of both geometry and link function. Real-world networks often violate these assumptions, exhibiting geometric mismatch and structural anomalies that break standard metric properties. We show that such misspecification pushes the data-generating distribution outside the model class, causing Bayesian inference to become overconfident and poorly calibrated. To address this, we propose a generalized posterior framework for random geometric graphs. We introduce Link-Sequential R-SafeBayes, a method that exploits dyadic conditional independence to estimate prequential risk and adaptively tune posterior regularization. Experiments on synthetic and real-world networks demonstrate improved calibration, better link prediction performance, and a reliable criterion for selecting latent geometries across Euclidean, spherical, and hyperbolic spaces.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The paper argues that Bayesian latent space models for graphs are misspecified when the assumed geometry or link function does not match the true data-generating process, leading to overconfident and poorly calibrated posteriors. It proposes a generalized posterior framework for random geometric graphs and introduces Link-Sequential R-SafeBayes, which exploits dyadic conditional independence to estimate prequential risk and adaptively tune the regularization parameter. Experiments on synthetic and real-world networks are claimed to show improved calibration, better link prediction, and a reliable way to select among Euclidean, spherical, and hyperbolic latent geometries.

Significance. If the adaptive regularization via prequential risk estimation proves reliable under misspecification, the work could offer a practical advance for robust network representation learning in statistics and machine learning. The idea of using sequential predictive risk to guide generalized posteriors is conceptually appealing and addresses a real issue in latent space models. Credit is due for focusing on geometry selection across multiple spaces and for providing experimental evidence of improved calibration, though the strength of these claims depends on resolving the theoretical properties of the risk estimator.

major comments (2)
  1. [Link-Sequential R-SafeBayes] Link-Sequential R-SafeBayes procedure (as described in the abstract and method): the adaptive tuning computes a sequential predictive risk by treating edges as conditionally independent given current latent positions and uses this to select the regularization parameter. When the true geometry differs from the assumed one (e.g., hyperbolic data fitted with Euclidean latents), the link probabilities entering the risk calculation are systematically incorrect; the resulting risk estimate can therefore be biased in a way that either over- or under-regularizes the posterior. No concentration inequality, bias bound, or robustness analysis for this estimator under model mismatch is provided, which is load-bearing for the central claim that the method restores calibration.
  2. [Abstract / Experiments] Abstract and experimental claims: the manuscript states that experiments demonstrate improved calibration and link prediction, yet provides no derivations, error analysis, or data-exclusion details for the risk estimator. Without these, it is difficult to verify that the reported gains are attributable to the adaptive procedure rather than other factors, undermining the link between the proposed method and the claimed improvements.
minor comments (2)
  1. [Introduction] Notation for the generalized posterior and the regularization parameter could be clarified with an explicit equation early in the manuscript to aid readability.
  2. [Abstract] The abstract mentions 'structural anomalies that break standard metric properties' but does not define these anomalies or link them explicitly to the misspecification analysis; a brief example or reference would help.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and insightful comments on our manuscript. We address each of the major comments in detail below and have revised the paper accordingly to improve clarity and strengthen the presentation of our results.

read point-by-point responses

  1. Referee: [Link-Sequential R-SafeBayes] Link-Sequential R-SafeBayes procedure (as described in the abstract and method): the adaptive tuning computes a sequential predictive risk by treating edges as conditionally independent given current latent positions and uses this to select the regularization parameter. When the true geometry differs from the assumed one (e.g., hyperbolic data fitted with Euclidean latents), the link probabilities entering the risk calculation are systematically incorrect; the resulting risk estimate can therefore be biased in a way that either over- or under-regularizes the posterior. No concentration inequality, bias bound, or robustness analysis for this estimator under model mismatch is provided, which is load-bearing for the central claim that the method restores calibration.

    Authors: We agree with the referee that the risk estimator can be biased under geometric misspecification because it relies on link probabilities computed from the assumed (potentially incorrect) geometry. This is a valid concern for the theoretical justification. In response, we have added a new subsection in the revised manuscript that provides a bias analysis for the prequential risk estimator under misspecification and includes additional simulation studies demonstrating the method's performance when the assumed geometry differs from the data-generating process. While a general concentration inequality that holds for arbitrary degrees of misspecification is not derived (as it would depend on quantifying the mismatch), we show that the adaptive regularization still leads to better calibrated posteriors in practice compared to standard Bayesian inference. revision: partial

  2. Referee: [Abstract / Experiments] Abstract and experimental claims: the manuscript states that experiments demonstrate improved calibration and link prediction, yet provides no derivations, error analysis, or data-exclusion details for the risk estimator. Without these, it is difficult to verify that the reported gains are attributable to the adaptive procedure rather than other factors, undermining the link between the proposed method and the claimed improvements.

    Authors: We thank the referee for pointing this out. To address this, the revised manuscript now includes a detailed derivation of the Link-Sequential R-SafeBayes estimator in the appendix, along with an analysis of the approximation error arising from the dyadic conditional independence assumption. We have also added explicit descriptions of the data splitting and exclusion procedures used in the real-world network experiments. These changes help establish that the reported improvements in calibration and link prediction can be attributed to the adaptive tuning of the regularization parameter. revision: yes

Circularity Check

0 steps flagged

No significant circularity; method introduces independent regularization tuning

full rationale

The paper proposes a generalized posterior for misspecified latent space models and introduces Link-Sequential R-SafeBayes to adaptively tune regularization by estimating prequential risk under dyadic conditional independence. This construction is presented as a new procedure whose validity is checked via experiments on synthetic and real networks rather than being forced by definition or prior self-citation. No load-bearing step reduces the central claim to a tautological fit or renamed input; the adaptive step is an explicit algorithmic choice whose performance is externally validated rather than guaranteed by construction. The derivation chain therefore remains self-contained against the stated assumptions and benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The abstract supplies almost no explicit free parameters or invented entities. The central proposal rests on the domain assumption of dyadic conditional independence and on the unstated claim that prequential risk can be estimated reliably from sequential link observations.

axioms (1)
  • domain assumption Dyadic conditional independence holds sufficiently well to allow sequential estimation of prequential risk.
    Invoked to justify the Link-Sequential R-SafeBayes procedure for tuning the generalized posterior.

pith-pipeline@v0.9.0 · 5665 in / 1276 out tokens · 40982 ms · 2026-05-20T08:34:32.444110+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

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

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

Reference graph

Works this paper leans on

62 extracted references · 62 canonical work pages

  1. [2]

    Journal of the american Statistical association , volume=

    Latent space approaches to social network analysis , author=. Journal of the american Statistical association , volume=. 2002 , publisher=

    work page 2002

  2. [3]

    2003 , publisher=

    Random geometric graphs , author=. 2003 , publisher=

    work page 2003

  3. [4]

    Physical Review E—Statistical, Nonlinear, and Soft Matter Physics , volume=

    Hyperbolic geometry of complex networks , author=. Physical Review E—Statistical, Nonlinear, and Soft Matter Physics , volume=. 2010 , publisher=

    work page 2010

  4. [5]

    Physical Review Research , volume=

    Link prediction with hyperbolic geometry , author=. Physical Review Research , volume=. 2020 , publisher=

    work page 2020

  5. [6]

    Proceedings of the Thirty-First Conference on Uncertainty in Artificial Intelligence , pages =

    Asta, Dena Marie and Shalizi, Cosma Rohilla , title =. Proceedings of the Thirty-First Conference on Uncertainty in Artificial Intelligence , pages =. 2015 , isbn =

    work page 2015

  6. [7]

    On the evolution of random graphs , author=. Publ. Math. Inst. Hungar. Acad. Sci , volume=

    work page

  7. [8]

    The Annals of Mathematical Statistics , volume=

    Random graphs , author=. The Annals of Mathematical Statistics , volume=. 1959 , publisher=

    work page 1959

  8. [9]

    Statistical science: a review journal of the Institute of Mathematical Statistics , volume=

    The geometry of continuous latent space models for network data , author=. Statistical science: a review journal of the Institute of Mathematical Statistics , volume=

    work page

  9. [10]

    International conference on machine learning , pages=

    Representation tradeoffs for hyperbolic embeddings , author=. International conference on machine learning , pages=. 2018 , organization=

    work page 2018

  10. [11]

    Journal of Physics: Complexity , volume=

    Zoo guide to network embedding , author=. Journal of Physics: Complexity , volume=. 2023 , publisher=

    work page 2023

  11. [12]

    IEEE transactions on knowledge and data engineering , volume=

    A comprehensive survey of graph embedding: Problems, techniques, and applications , author=. IEEE transactions on knowledge and data engineering , volume=. 2018 , publisher=

    work page 2018

  12. [13]

    APS March meeting abstracts , volume=

    Clustering means geometry in networks , author=. APS March meeting abstracts , volume=

    work page

  13. [14]

    Scientific reports , volume=

    Emergence of soft communities from geometric preferential attachment , author=. Scientific reports , volume=. 2015 , publisher=

    work page 2015

  14. [15]

    Physical Review E—Statistical, Nonlinear, and Soft Matter Physics , volume=

    Curvature and temperature of complex networks , author=. Physical Review E—Statistical, Nonlinear, and Soft Matter Physics , volume=. 2009 , publisher=

    work page 2009

  15. [16]

    Discrete & Computational Geometry , volume=

    Ollivier’s Ricci curvature, local clustering and curvature-dimension inequalities on graphs , author=. Discrete & Computational Geometry , volume=. 2014 , publisher=

    work page 2014

  16. [17]

    Tohoku Mathematical Journal, Second Series , volume=

    Ricci curvature of graphs , author=. Tohoku Mathematical Journal, Second Series , volume=. 2011 , publisher=

    work page 2011

  17. [18]

    PloS one , volume=

    Embedding graphs in Lorentzian spacetime , author=. PloS one , volume=. 2017 , publisher=

    work page 2017

  18. [19]

    Proceedings of the thirtieth annual symposium on Computational geometry , pages=

    Metric embedding, hyperbolic space, and social networks , author=. Proceedings of the thirtieth annual symposium on Computational geometry , pages=

    work page

  19. [20]

    Journal of Complex Networks , volume=

    Hydra: a method for strain-minimizing hyperbolic embedding of network-and distance-based data , author=. Journal of Complex Networks , volume=. 2020 , publisher=

    work page 2020

  20. [21]

    Journal of Complex Networks , volume=

    An improved hyperbolic embedding algorithm , author=. Journal of Complex Networks , volume=. 2018 , publisher=

    work page 2018

  21. [22]

    Physical Review E , volume=

    Systematic comparison of graph embedding methods in practical tasks , author=. Physical Review E , volume=. 2021 , publisher=

    work page 2021

  22. [23]

    Communications Physics , volume=

    Symmetry-driven embedding of networks in hyperbolic space , author=. Communications Physics , volume=. 2025 , publisher=

    work page 2025

  23. [24]

    Nature Physics , volume=

    Network structure from rich but noisy data , author=. Nature Physics , volume=. 2018 , publisher=

    work page 2018

  24. [25]

    Journal of Complex Networks , volume=

    Bayesian inference of network structure from unreliable data , author=. Journal of Complex Networks , volume=. 2020 , publisher=

    work page 2020

  25. [26]

    Physical Review X , volume=

    Reconstructing networks with unknown and heterogeneous errors , author=. Physical Review X , volume=. 2018 , publisher=

    work page 2018

  26. [27]

    S. L. van der Pas and A. W. van der Vaart , title =. Bayesian Analysis , number =. 2018 , doi =

    work page 2018

  27. [28]

    Journal of computational and graphical statistics , volume=

    Fast inference for the latent space network model using a case-control approximate likelihood , author=. Journal of computational and graphical statistics , volume=. 2012 , publisher=

    work page 2012

  28. [29]

    International Colloquium on Automata, Languages, and Programming , pages=

    Random hyperbolic graphs: degree sequence and clustering , author=. International Colloquium on Automata, Languages, and Programming , pages=. 2012 , organization=

    work page 2012

  29. [30]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    Identifying the latent space geometry of network models through analysis of curvature , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2023 , publisher=

    work page 2023

  30. [31]

    Bayesian Analysis , number =

    Peter Gr. Bayesian Analysis , number =. 2017 , doi =

    work page 2017

  31. [32]

    Journal of the Royal Statistical Society: Series A (General) , volume=

    Present position and potential developments: Some personal views statistical theory the prequential approach , author=. Journal of the Royal Statistical Society: Series A (General) , volume=. 1984 , publisher=

    work page 1984

  32. [33]

    2013 , booktitle =

    KONECT - The Koblenz Network Collection , author =. 2013 , booktitle =

    work page 2013

  33. [34]

    D. M. Asta and C. R. Shalizi. Geometric network comparisons. In Proceedings of the Thirty-First Conference on Uncertainty in Artificial Intelligence, UAI'15, page 102–110, Arlington, Virginia, USA, 2015. AUAI Press. ISBN 9780996643108

    work page 2015

  34. [35]

    Chowdhary and T

    K. Chowdhary and T. G. Kolda. An improved hyperbolic embedding algorithm. Journal of Complex Networks, 6 0 (3): 0 321--341, 2018

    work page 2018

  35. [36]

    J. R. Clough and T. S. Evans. Embedding graphs in lorentzian spacetime. PloS one, 12 0 (11): 0 e0187301, 2017

    work page 2017

  36. [37]

    A. P. Dawid. Present position and potential developments: Some personal views statistical theory the prequential approach. Journal of the Royal Statistical Society: Series A (General), 147 0 (2): 0 278--290, 1984

    work page 1984

  37. [38]

    Erd o s and A

    P. Erd o s and A. R \'e nyi. On the evolution of random graphs. Publ. Math. Inst. Hungar. Acad. Sci, 5 0 (1): 0 17--61, 1960

    work page 1960

  38. [39]

    E. N. Gilbert. Random graphs. The Annals of Mathematical Statistics, 30 0 (4): 0 1141--1144, 1959

    work page 1959

  39. [40]

    Gutmann, M

    P. Gr \"u nwald and T. van Ommen. Inconsistency of Bayesian Inference for Misspecified Linear Models, and a Proposal for Repairing It . Bayesian Analysis, 12 0 (4): 0 1069 -- 1103, 2017. doi:10.1214/17-BA1085. URL https://doi.org/10.1214/17-BA1085

    work page doi:10.1214/17-ba1085 2017

  40. [41]

    Gugelmann, K

    L. Gugelmann, K. Panagiotou, and U. Peter. Random hyperbolic graphs: degree sequence and clustering. In International Colloquium on Automata, Languages, and Programming, pages 573--585. Springer, 2012

    work page 2012

  41. [42]

    P. D. Hoff, A. E. Raftery, and M. S. Handcock. Latent space approaches to social network analysis. Journal of the american Statistical association, 97 0 (460): 0 1090--1098, 2002

    work page 2002

  42. [43]

    Jost and S

    J. Jost and S. Liu. Ollivier’s ricci curvature, local clustering and curvature-dimension inequalities on graphs. Discrete & Computational Geometry, 51 0 (2): 0 300--322, 2014

    work page 2014

  43. [44]

    Keller-Ressel and S

    M. Keller-Ressel and S. Nargang. Hydra: a method for strain-minimizing hyperbolic embedding of network-and distance-based data. Journal of Complex Networks, 8 0 (1): 0 cnaa002, 2020

    work page 2020

  44. [45]

    Kitsak, I

    M. Kitsak, I. Voitalov, and D. Krioukov. Link prediction with hyperbolic geometry. Physical Review Research, 2 0 (4): 0 043113, 2020

    work page 2020

  45. [46]

    Krioukov

    D. Krioukov. Clustering means geometry in networks. In APS March meeting abstracts, volume 2016, pages Y12--005, 2016

    work page 2016

  46. [47]

    Krioukov, F

    D. Krioukov, F. Papadopoulos, A. Vahdat, and M. Bogun \'a . Curvature and temperature of complex networks. Physical Review E—Statistical, Nonlinear, and Soft Matter Physics, 80 0 (3): 0 035101, 2009

    work page 2009

  47. [48]

    Krioukov, F

    D. Krioukov, F. Papadopoulos, M. Kitsak, A. Vahdat, and M. Bogun \'a . Hyperbolic geometry of complex networks. Physical Review E—Statistical, Nonlinear, and Soft Matter Physics, 82 0 (3): 0 036106, 2010

    work page 2010

  48. [49]

    J. Kunegis. Konect - the koblenz network collection. In Proc. Int. Conf. on World Wide Web Companion, pages 1343--1350, 2013

    work page 2013

  49. [50]

    Y. Lin, L. Lu, and S.-T. Yau. Ricci curvature of graphs. Tohoku Mathematical Journal, Second Series, 63 0 (4): 0 605--627, 2011

    work page 2011

  50. [51]

    Lizotte, J.-G

    S. Lizotte, J.-G. Young, and A. Allard. Symmetry-driven embedding of networks in hyperbolic space. Communications Physics, 8 0 (1): 0 199, 2025

    work page 2025

  51. [52]

    Lubold, A

    S. Lubold, A. G. Chandrasekhar, and T. H. McCormick. Identifying the latent space geometry of network models through analysis of curvature. Journal of the Royal Statistical Society Series B: Statistical Methodology, 85 0 (2): 0 240--292, 2023

    work page 2023

  52. [53]

    M. E. Newman. Network structure from rich but noisy data. Nature Physics, 14 0 (6): 0 542--545, 2018

    work page 2018

  53. [54]

    Papamichalis, K

    M. Papamichalis, K. Turnbull, S. Lunagomez, and E. Airoldi. Latent space network modelling with hyperbolic and spherical geometries. arXiv preprint arXiv:2109.03343, 2021

    work page arXiv 2021

  54. [55]

    T. P. Peixoto. Reconstructing networks with unknown and heterogeneous errors. Physical Review X, 8 0 (4): 0 041011, 2018

    work page 2018

  55. [56]

    M. Penrose. Random geometric graphs, volume 5. OUP Oxford, 2003

    work page 2003

  56. [57]

    A. E. Raftery, X. Niu, P. D. Hoff, and K. Y. Yeung. Fast inference for the latent space network model using a case-control approximate likelihood. Journal of computational and graphical statistics, 21 0 (4): 0 901--919, 2012

    work page 2012

  57. [58]

    F. Sala, C. De Sa, A. Gu, and C. R \'e . Representation tradeoffs for hyperbolic embeddings. In International conference on machine learning, pages 4460--4469. PMLR, 2018

    work page 2018

  58. [59]

    A. L. Smith, D. M. Asta, and C. A. Calder. The geometry of continuous latent space models for network data. Statistical science: a review journal of the Institute of Mathematical Statistics, 34 0 (3): 0 428, 2019

    work page 2019

  59. [60]

    S. L. van der Pas and A. W. van der Vaart. Bayesian Community Detection . Bayesian Analysis, 13 0 (3): 0 767 -- 796, 2018. doi:10.1214/17-BA1078. URL https://doi.org/10.1214/17-BA1078

    work page doi:10.1214/17-ba1078 2018

  60. [61]

    Verbeek and S

    K. Verbeek and S. Suri. Metric embedding, hyperbolic space, and social networks. In Proceedings of the thirtieth annual symposium on Computational geometry, pages 501--510, 2014

    work page 2014

  61. [62]

    Young, G

    J.-G. Young, G. T. Cantwell, and M. Newman. Bayesian inference of network structure from unreliable data. Journal of Complex Networks, 8 0 (6): 0 cnaa046, 2020

    work page 2020

  62. [63]

    K. Zuev, M. Bogu \ n \'a , G. Bianconi, and D. Krioukov. Emergence of soft communities from geometric preferential attachment. Scientific reports, 5 0 (1): 0 9421, 2015

    work page 2015