Pith. sign in

REVIEW 4 major objections 5 minor 50 references

Graph statistics: An emerging discipline in non-Euclidean data analysis

T0 review · 4 major / 5 minor · reviewed 2026-07-15 · grok-4.5

Pith's one-line read Graph statistics treats modern data as dynamic networks, not Euclidean vectors, and rebuilds them from static samples via quasi-dynamic models.

desk verdict Readable manifesto packaging the authors' prior idopNet/qdMODE/GLMY pipeline as a 'new discipline'; no new theorem or independent check of the static-to-dynamic hinge. read the letter →

arxiv 2607.10324 v2 pith:RGEMHFBD submitted 2026-07-11 stat.ME

classification stat.ME MSC 62H9962P1005C9037N25
keywords graphstatisticsnon-EuclideandataidopNetquasi-dynamicmodelingevolutionarygametheoryallometricscalingGLMYhomologymultilayermodularnetworks
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 argues that classical statistics, built for independent points in Euclidean space, cannot handle the relational data that dominate modern science—gene circuits, microbial webs, neural circuits, materials, and AI graphs. It presents graph statistics as a unified response: reconstruct multilayer modular sparse networks from high-dimensional measurements, recover directed, signed, and weighted interactions for each sample (idopNet), and map topological features of those networks to function and genetics. The key technical move is to convert static cross-sectional panels into quasi-dynamic trajectories using ecological niche theory and allometric scaling, then solve a system of mixed ordinary differential equations by multi-task learning that jointly performs variable selection and interaction estimation. Evolutionary game theory supplies the formation rules that explain hubs, motifs, and modularity; GLMY homology supplies the topological invariants that link architecture to behavior. The authors claim this pipeline yields mechanistic, sample-specific networks that can be treated as phenotypes and that can give graph neural networks causal structure rather than black-box aggregation.

What carries the argument

idopNet reconstructed by quasi-dynamic mixed ordinary differential equations (qdMODE): each agent’s trajectory is split into an intrinsic term plus the cumulative influence of all others; multi-task double-sparse learning recovers directed, signed, weighted edges from static samples treated as points along an ecological/allometric gradient.

What would settle it

On a system with known ground-truth directed interactions and dense time series (e.g., a controlled microbial cross-feeding community or a simulated gene circuit), reconstruct idopNets from deliberately sparsified static snapshots ordered only by allometric or niche proxies; if the recovered edge signs, directions, and hub structure systematically disagree with the true dynamics, the central reconstruction claim fails.

Watch

Extended reading notes

Core claim

Graph statistics is a coherent non-Euclidean statistical discipline that reconstructs informative, dynamic, omnidirectional, personalized networks (idopNet) from static high-dimensional data by decomposing each agent’s change into independent and dependent components via quasi-dynamic mixed ODEs, multi-task sparse learning, evolutionary game theory, and topological (GLMY) dissection, thereby turning network architecture itself into a mappable phenotype.

Load-bearing premise

Static cross-sectional samples can be ordered by ecological niche and allometric scaling so that the resulting quasi-dynamic equations recover the true independent and interaction terms of the underlying dynamical system.

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

4 major / 5 minor

Summary. This manuscript is a perspective/overview that proposes “graph statistics” as an emerging discipline for non-Euclidean, network-structured data. It contrasts Euclidean vector methods with relational data, then assembles a pipeline: evolutionary game theory as a generative account of network architecture; multilayer modular sparse reconstruction; multi-task learning of idopNet via a mixed ODE decomposition into independent and dependent terms, converted from static panels into quasi-dynamic MODE (qdMODE) using ecological niche theory and allometric scaling; GLMY topological dissection of the resulting digraphs; and claimed mechanistic interpretability gains for graph neural networks. Applications are sketched in quantitative genetics, systems biology, microbial ecology, materials science, and AI. The piece is largely synthetic and cites the authors’ prior work for the technical pillars rather than deriving or re-validating them here.

Significance. If the integrated pipeline works as claimed, the paper would usefully name and organize a coherent non-Euclidean statistical program that treats network topology itself as a phenotype and links reconstruction to dynamical and topological interpretation. The framing is timely for systems biology, microbiome analysis, and interpretable GNNs. Strengths include a clear Euclidean/non-Euclidean distinction, an explicit agent-level decomposition into independent and interaction terms, and a consistent multi-scale story from reconstruction through topology to AI. As a perspective, however, significance rests on synthesis and scope rather than new theorems, error bounds, or controlled comparisons supplied in this text; the revolutionary claims therefore depend on the fidelity of the static-to-dynamic bridge and related prior results.

major comments (4)
  1. The central causal/dynamical claim of the framework rests on converting static cross-sectional panels into quasi-dynamic trajectories (qdMODE) via ecological niche theory and allometric scaling (Multi-task learning of idopNet section; citations Griffin et al. 2020; He & Liu 2009). That step is load-bearing for directed signed edges, multi-task double-sparse estimation, GLMY mapping, and GNN interpretability. The manuscript supplies no new derivation, recovery bounds, or simulation quantifying fidelity under uneven gradients, non-allometric traits, or measurement noise. For a perspective that frames a “new discipline,” the paper should either (i) restate the precise conditions under which independent term Q_j(y_j) and interaction terms Q_jj'(y_j') are recovered with controlled error, or (ii) clearly demarcate which claims remain conditional on those external results.
  2. The multi-task double-sparse hard-thresholding procedure and the additive MODE decomposition (equation in Multi-task learning of idopNet) are presented as solving variable selection and equation solving jointly, yet free parameters (sparsity/threshold levels, module count or clustering resolution) and consistency guarantees are not stated in this text. Without even a brief statement of what is assumed versus what is proven in the cited Dong et al. (2026) / Liu et al. (2026) work, the reader cannot assess whether the estimator is a principled solution or a tunable heuristic. A short, self-contained statement of assumptions and known recovery conditions is needed.
  3. GLMY dissection is asserted to identify topological features that “govern system behavior” and enable prediction of metabolic phenotypes or community stability from structure alone (GLMY dissection of idopNet). The manuscript does not specify which invariants are used as predictors, how they are linked statistically to outcomes, or what validation (cross-system, hold-out, or perturbation) supports predictive claims. Either add a concrete mapping (feature → phenotype) with citation to quantitative results, or tone the language from prediction to descriptive characterization.
  4. The section “How graph statistics enables graph neural networks” claims that the framework supplies mathematical invariants, mechanistic causality, and geometric constraints that move GNNs from black-box prediction to rule-based structural discovery. No architecture, loss, theorem, or empirical comparison is given. For the claim to remain load-bearing rather than aspirational, the paper should either cite a concrete IdopGNN-style result with a measurable interpretability or robustness gain, or reframe the section as a research program rather than an established enablement.
minor comments (5)
  1. Figure 1 is described in the caption but the main text does not walk the reader through panels (a)–(e) in order; a short paragraph tying the figure to the idopNet vs. correlation/Bayesian contrast would help.
  2. Notation for the mixed ODE uses Q_j and Q_jj' without stating regularity or identifiability conditions; a one-sentence clarification would reduce ambiguity.
  3. Several key citations are listed as 2026 (Dong et al.; Sun et al.; Ma et al.; Xu et al.). For a journal version, status (in press / published / arXiv) should be made explicit so readers can access the technical backbone.
  4. The keywords list “allometric scaling law” and “idopNet” but the abstract does not mention allometry; aligning abstract and keywords would improve discoverability.
  5. Occasional phrasing (“new norm of statistical thinking,” “poised to revolutionize”) is stronger than the evidence presented in this overview; tempering in the abstract and conclusion would better match the perspective genre.

Circularity Check

3 steps flagged · score 4.0 of 10

Overview paper whose methodological pillars rest on load-bearing self-citations, but with no within-paper prediction that reduces to its inputs by construction.

  1. self citation load bearing [Multi-task learning of idopNet (qdMODE construction)]
    "To overcome this limitation, ecological niche theory is introduced to extract dynamic information from static data [8]. The allometric scaling law, which relates organismal traits to body size across species, provides a principled basis for inferring dynamic trajectories from cross-sectional observations [33]. By treating samples as points along an inferred temporal or ecological gradient, a system of quasi-dynamic MODE (qdMODE) is constructed. ... The theoretical properties of this quasi-dynamic extension have been established and validated [34]."

    The static-to-dynamic bridge that gives idopNet its dynamical/causal reading is not derived or error-bounded in this manuscript. Validation is deferred entirely to [8] (Wu & Jiang) and [34] (Griffin, Jiang & Wu)—overlapping authors—so the overview’s claim that qdMODE recovers independent and dependent interaction terms rests on accepting the authors’ prior results as given rather than on independent support presented here.

  2. self citation load bearing [Multi-task learning of idopNet (empirical utility paragraph)]
    "The practical utility of this framework is demonstrated through application to gene regulatory network reconstruction for the malaria parasite Plasmodium falciparum [32]. Using transcriptional data from infection time courses, the multi-task learning model identifies previously unknown regulatory roles of several genes in mediating malaria infection."

    The sole concrete demonstration that multi-task idopNet yields novel biological insight is citation [32] (Dong, Fa, Li, Yau & Wu)—the same author group. The overview’s claim that the framework generates mechanistic discoveries therefore reduces to self-citation rather than external evidence.

1 more flagged steps
  1. self citation load bearing [GLMY dissection of idopNet (applications)]
    "The application of GLMY dissection to metabolic networks illustrates its biological utility [9]. ... Microbial interaction networks present a distinct but equally powerful application domain [10,14]. ... Beyond biology, the GLMY framework finds natural application in materials science ... [38]."

    Illustrative applications that are offered as evidence the topological mapping works are almost entirely the authors’ own prior papers ([9], [10], [14]; GLMY foundations [36] include Yau). The overview’s assertion that GLMY connects network topology to function therefore leans on the same research program it is promoting, without independent external benchmarks in this text.

full rationale

This manuscript is a perspective/overview of a research program, not a first-principles derivation that claims to predict observables from independent axioms. Euclidean vs non-Euclidean distinctions, modularity, sparsity, and evolutionary game theory are argued with external citations (Newman, Maynard Smith, etc.) and have independent content. The load-bearing dynamical claim—that static panels can be converted into quasi-dynamic MODE trajectories that recover independent and dependent interaction terms—is not re-derived here; it is imported from prior work by overlapping authors (Wu & Jiang 2021; Griffin, Jiang & Wu 2020) and treated as established. Empirical ‘demonstrations’ (malaria GRN, metabolic/microbial GLMY applications) likewise point only to the authors’ own papers. That is self-citation load-bearing for the ‘new discipline’ framing, but it is not a fitted-input-called-prediction or self-definitional loop inside this text: no parameter is fit to data in this paper and then re-presented as a prediction, and no uniqueness theorem is invoked to forbid alternatives. Per the rules, heavy self-citation in an overview of one’s own program is not automatic circularity; score 4 reflects partial dependence of the central methodological narrative on unverified-here self-citations without a by-construction reduction of a claimed result.

Assumptions & free parameters 2 free parameters · 4 assumptions · 3 invented entities

As a review/manifesto the paper contributes almost no new free parameters or axioms of its own; the load-bearing content is imported from the authors' prior program. The ledger therefore records the domain assumptions and invented constructs that the overview treats as settled foundations for the claimed new discipline.

free parameters (2)
  • sparsity / hard-threshold levels in multi-task double-sparse estimation
    Network edge selection depends on group- and element-wise thresholds whose values are not fixed by theory in this text and must be chosen or tuned in the underlying algorithms.
  • module / community count or clustering resolution
    Multilayer modular reconstruction begins with clustering into modules; the number or resolution of modules is a modeling choice that shapes the recovered graph.
assumptions (4)
  • domain assumption Static cross-sectional samples ordered by an ecological or allometric gradient yield quasi-dynamic trajectories that preserve the independent and dependent terms of the true mixed ODE system.
    Stated as the bridge that allows idopNet reconstruction without dense time series (Multi-task learning of idopNet; cites Griffin et al. 2020).
  • domain assumption Evolutionary game-theoretic incentives (cooperation, exploitation, mutualism) are the generative mechanism that produces observed biological network motifs and hubs.
    Used throughout 'Evolutionary game theory in action' to explain and predict network architecture.
  • domain assumption GLMY digraph homology invariants (sources, sinks, cycles, hierarchy) encode the dynamical and functional properties of the system.
    Core of the GLMY dissection section; imported from Grigor'yan–Lin–Muranov–Yau path homology and subsequent applications.
  • ad hoc to paper Agent dynamics decompose additively into an independent self-term Q_j(y_j) plus sum of dependent interaction terms Q_jj'(y_j').
    Defining equation of idopNet; the additive MODE form is a modeling choice that enables the multi-task matrix estimation.
invented entities (3)
  • idopNet (informative, dynamic, omnidirectional, personalized network)
    purpose: Named target object of reconstruction: sample-specific signed directed weighted networks from the MODE decomposition.
    Central construct of the authors' program; independent evidence is limited to applications in their own prior papers.
  • qdMODE (quasi-dynamic mixed ordinary differential equations)
    purpose: Replace true time derivatives with sample-ordered trajectories so static panels can be treated as dynamic systems.
    Enables the whole static-to-network pipeline; validation is internal to the authors' citations.
  • graph statistics (as a named discipline)
    purpose: Umbrella label that unifies the authors' network reconstruction, game-theoretic, and topological tools as a successor to Euclidean statistics.
    Rhetorical and organizational invention of this overview; not an independently falsifiable object.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Graph statistics: An emerging discipline in non-Euclidean data analysis." pith.science (2026). https://pith.science/paper/RGEMHFBD

@misc{pith2026260710324,
  author       = {Pith},
  title        = {Pith review of: Graph statistics: An emerging discipline in non-Euclidean data analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RGEMHFBD}},
  note         = {Machine review of arXiv:2607.10324}
}
read the original abstract

The explosive growth of complex data has catalyzed the emergence of graph statistics as a fundamentally new discipline in data science. Unlike traditional statistics, which operates primarily within the comfortable confines of Euclidean spaces, graph statistics confronts the reality that modern data naturally organize themselves as dynamic networks composed of complex interconnections. In this article, we present an overview of graph statistics as an emerging discipline, tracing its theoretical foundations, methodological innovations, and transformative applications. We examine how the integration of evolutionary game theory, ecological niche theory, topological data analysis, and graph theory through quasi-dynamic nonlinear modeling has created a new norm of statistical thinking capable of analyzing non-Euclidean data. We show how graph statistics is poised to revolutionize fields ranging from quantitative genetics and systems biology to materials science and artificial intelligence, offering a principled framework for transforming big data into practical knowledge.

Figures

Figures reproduced from arXiv: 2607.10324 by the authors.

Figure 1
Figure 1. Reconstructing idopNet: From static data to causative networks. [PITH_FULL_IMAGE:figures/full_fig_p016_1.png] view at source ↗
Figure 1
Figure 1. Reconstructing idopNetwork: From static data to causative networks. (a) Sample collection: A panel of n studied objects are sampled from 1 to n [PITH_FULL_IMAGE:figures/full_fig_p014_1.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

50 extracted references · 2 linked inside Pith

  1. [1]

    E., Searle, S

    McCulloch, C. E., Searle, S. R., & Neuhaus, J. M. (2008). Generalized, linear, and mixed models. John Wiley & Sons

  2. [2]

    Christensen, R. (2018). Analysis of variance, design, and regression: Linear modeling for unbalanced data. Chapman and Hall/CRC

  3. [3]

    Zhang, K., Liu, S., & Xiong, M. (2022). Changes from classical statistics to modern statistics and data science. arXiv preprint arXiv:2211.03756

  4. [4]

    Martín, N. (2025). Data Science Versus Statistics. In International Encyclopedia of Statistical Science (pp. 621-623). Berlin, Heidelberg: Springer Berlin Heidelberg

  5. [5]

    Carthew, R. W. (2021). Gene regulation and cellular metabolism: an essential partnership. Trends in Genetics, 37(4), 389-400

  6. [6]

    Luo, L. (2021). Architectures of neuronal circuits. Science, 373(6559), eabg7285

  7. [7]

    Barraclough, T. G. (2015). How do species interactions affect evolutionary dynamics across whole communities? Annual Review of Ecology, Evolution, and Systematics, 46(1), 25-48

  8. [8]

    Wu, R., & Jiang, L. (2021). Recovering dynamic networks in big static datasets. Physics Reports, 912, 1-57

Show all 50 references
  1. [9]

    Wu, S., Liu, X., Dong, A., Gragnoli, C., Griffin, C., Wu, J., ... & Wu, R. (2023). The metabolomic physics of complex diseases. Proceedings of the National Academy of Sciences, 120(42), e2308496120

  2. [10]

    Feng, L., Gong, H., Zhang, S., Liu, X., Wang, Y ., Che, J., ... & Wu, R. (2024). Hypernetwork modeling and topology of high-order interactions for complex systems. Proceedings of the National Academy of Sciences, 121(40), e2412220121

  3. [11]

    Sun, L., Bian, Y ., Yang, D., Miao, R., Meng, Y ., Che, J., ... & Wu, R. (2026). Graph statistics theory of individualized quantitative genetics under haplotype-resolved genome assembly. Proceedings of the National Academy of Sciences, 123(14), e2600004123

  4. [12]

    Chen, C., Jiang, L., Fu, G., Wang, M., Wang, Y ., Shen, B., ... & Wu, R. (2019). An omnidirectional visualization model of personalized gene regulatory networks. NPJ systems biology and applications, 5(1), 38

  5. [13]

    Dong, A., Wu, S., Che, J., Wang, Y ., & Wu, R. (2023). idopNetwork: a network tool to dissect spatial community ecology. Methods in Ecology and Evolution, 14(9), 2272-2283

  6. [14]

    Gong, H., Wang, H., Wang, Y ., Zhang, S., Liu, X., Che, J., ... & Wu, R. (2024). Topological change of soil microbiota networks for forest resilience under global warming. Physics of Life 12 Reviews, 50, 228-251

  7. [15]

    E., Tang, M., Priebe, C

    Athreya, A., Fishkind, D. E., Tang, M., Priebe, C. E., Park, Y ., V ogelstein, J. T., ... & Sussman, D. L. (2018). Statistical inference on random dot product graphs: a survey. Journal of Machine Learning Research, 18(226), 1-92

  8. [16]

    (2019, April)

    Charikar, M., Chatziafratis, V ., Niazadeh, R., & Yaroslavtsev, G. (2019, April). Hierarchical clustering for euclidean data. In The 22nd International Conference on Artificial Intelligence and Statistics (pp. 2721-2730). PMLR

  9. [17]

    Song, W., Zhou, H., Zhou, Y ., & Müller, H. G. (2026). Non-Euclidean data analysis with metric statistics. Harvard Data Science Review, 8

  10. [18]

    Faraway, J. J. (2014). Regression for non-Euclidean data using distance matrices. Journal of Applied Statistics, 41(11), 2342-2357

  11. [19]

    (2022, August)

    Yang, J., Xie, K., & An, N. (2022, August). Causal discovery on non-euclidean data. In Proceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining (pp. 2202-2211)

  12. [20]

    M., & Price, G

    Smith, J. M., & Price, G. R. (1973). The logic of animal conflict. Nature, 246(5427), 15-18

  13. [21]

    Su, Q., McAvoy, A., & Plotkin, J. B. (2023). Strategy evolution on dynamic networks. Nature Computational Science, 3(9), 763-776

  14. [22]

    Sheng, A., Su, Q., Wang, L., & Plotkin, J. B. (2024). Strategy evolution on higher-order networks. Nature computational science, 4(4), 274-284

  15. [23]

    H., Huang, L

    Yeang, C. H., Huang, L. C., & Liu, W. C. (2012, August). Recurrent structural motifs reflect characteristics of distinct networks. In 2012 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (pp. 551-557). IEEE

  16. [24]

    Newman, M. E. (2006). Modularity and community structure in networks. Proceedings of the national academy of sciences, 103(23), 8577-8582

  17. [25]

    Ghavasieh, A., & De Domenico, M. (2024). Diversity of information pathways drives sparsity in real-world networks. Nature Physics, 20(3), 512-519

  18. [26]

    A., & Bioucas-Dias, J

    Figueiredo, M. A., & Bioucas-Dias, J. M. (2012, May). Algorithms for imaging inverse problems under sparsity regularization. In 2012 3rd International Workshop on Cognitive Information Processing (CIP) (pp. 1-6). IEEE

  19. [27]

    Bach, F., Jenatton, R., Mairal, J., & Obozinski, G. (2012). Optimization with sparsity-inducing penalties. Foundations and Trends in Machine Learning, 4(1), 1-106

  20. [28]

    R., Zhang, L., Berg, A., Fan, J., & Wu, R

    Kim, B. R., Zhang, L., Berg, A., Fan, J., & Wu, R. (2008). A computational approach to the functional clustering of periodic gene-expression profiles. Genetics, 180(2), 821-834

  21. [29]

    Wang, Y ., Xu, M., Wang, Z., Tao, M., Zhu, J., Wang, L., ... & Wu, R. (2012). How to cluster gene expression dynamics in response to environmental signals. Briefings in bioinformatics, 13(2), 162-174

  22. [30]

    Pan, W., Che, J., & Wu, S. (2026). An improved method of Wu's functional clustering. Statistics Innovation, 3(1), e005

  23. [31]

    George, E. I. (2000). The variable selection problem. Journal of the American Statistical Association, 95(452), 1304-1308

  24. [32]

    T., & Wu, R

    Dong, A., Fa, C., Li, Z., Yau, S. T., & Wu, R. (2026). Multi-task learning of complex networks via nonlinear ordinary differential equations. Communications Physics

  25. [33]

    H., & Liu, J

    He, J. H., & Liu, J. F. (2009). Allometric scaling laws in biology and physics. Chaos, Solitons & Fractals, 41(4), 1836-1838

  26. [34]

    Griffin, C., Jiang, L., & Wu, R. (2020). Analysis of quasi-dynamic ordinary differential equations and the quasi-dynamic replicator. Physica A: Statistical Mechanics and its Applications, 555, 124422

  27. [35]

    Liu, S., Li, Z., Zhang, Y ., & Yin, J. (2026). Exact recovery in the double sparse model: Sufficient and necessary signal conditions. Electronic Journal of Statistics, 20(1), 83-137. 13

  28. [36]

    Grigor'yan, A., Lin, Y ., Muranov, Y ., & Yau, S. T. (2012). Homologies of path complexes and digraphs. arXiv preprint arXiv:1207.2834

  29. [37]

    Wu, S., & Zhang, M. (2025). Disentangling complex systems: IdopNetwork meets GLMY homology theory. Data Analytics and Topology, 1(2025), 47-64

  30. [38]

    Wang, B., Feng, B., Lv, L., Li, S., & Pan, F. (2025). Structural feature extraction via topological data analysis. The Journal of Physical Chemistry Letters, 16(32), 8056-8067

  31. [39]

    C., Vidhyapriya, P., & Arun, V

    Vetrivel, S. C., Vidhyapriya, P., & Arun, V . P. (2025). The Challenges of Graph Neural Networks. In Graph Neural Networks: Essentials and Use Cases (pp. 79-108). Cham: Springer Nature Switzerland

  32. [40]

    Zednik, C. (2021). Solving the black box problem: A normative framework for explainable artificial intelligence. Philosophy & technology, 34(2), 265-288

  33. [41]

    & Miolane, N

    Papillon, M., Sanborn, S., Mathe, J., Cornelis, L., Bertics, A., Buracas, D., ... & Miolane, N. (2025). Beyond euclid: An illustrated guide to modern machine learning with geometric, topological, and algebraic structures. Machine Learning: Science and Technology, 6(3), 031002

  34. [42]

    Xiao, S., Wang, S., Dai, Y ., & Guo, W. (2022). Graph neural networks in node classification: survey and evaluation. Machine Vision and Applications, 33(1), 4

  35. [43]

    F., & Capretz, M

    L’heureux, A., Grolinger, K., Elyamany, H. F., & Capretz, M. A. (2017). Machine learning with big data: Challenges and approaches. IEEE Access, 5, 7776-7797

  36. [44]

    (2026) IdopGNN: A graph statistical mechanics model of data prediction

    Ma, S., Dong, A., Zhou G., Yang, F., Long, F., Wang, Y ., Yau, S.-T., Wu, R. (2026) IdopGNN: A graph statistical mechanics model of data prediction. Data Analytics and Topology (in press)

  37. [45]

    T., & Yau, S

    Yau, S. T., & Yau, S. S. T. (2000). Real time solution of nonlinear filtering problem without memory I. Mathematical Research Letters, 7(6), 671-693

  38. [46]

    T., & Yau, S

    Yau, S. T., & Yau, S. S. T. (2008). Real time solution of the nonlinear filtering problem without memory II. SIAM Journal on Control and Optimization, 47(1), 163-195

  39. [47]

    Wang, Y ., Xu, S., Wei, X., Luo, X., Yau, S. S. T., Yau, S. T., & Wu, R. (2025). Yau-YauAL: A computer tool for solving nonlinear filtering problems. Data Analytics and Topology, 1 (2025), 77-84

  40. [48]

    Xu, S., Wang, Y ., Wu, S., Dong, A., Yau, S. S. T., Yau, S. T., & Wu, R. (2026). Statistical learning of stochastic complex systems via the Yau-Yau nonlinear filter. The Innovation, 7(6), 101267

  41. [49]

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

  42. [50]

    Opper, M., & Saad, D. (Eds.). (2001). Advanced mean field methods: Theory and practice. MIT press. 14 Figure 1. Reconstructing idopNetwork: From static data to causative networks. (a) Sample collection: A panel of n studied objects are sampled from 1 to n. 15 (b) Agent-based d...

Pith tools

Reviewed July 15, 2026 · model on record in the stance chip above.