pith. sign in

arxiv: 2606.23741 · v1 · pith:VLJ2YWDEnew · submitted 2026-06-21 · 💻 cs.LG · cs.AI· cs.DC

A Survey on Federated Causal Discovery and Inference

Xianjie Guo , Yuwei Wang , Guodu Xiang , Xiaoli Tang , Kui Yu , Han Yu , Qiang Yang This is my paper

Pith reviewed 2026-06-26 10:21 UTC · model grok-4.3

classification 💻 cs.LG cs.AIcs.DC
keywords federated learningcausal discoverycausal inferenceprivacytaxonomydistributed datastructural learningtreatment effects
0
0 comments X

The pith

Federated causal discovery and inference methods are classified by three design decisions on learning, partitioning, and knowledge sharing.

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

This survey supplies a map for the emerging area of federated causal discovery and inference, where institutions collaborate on causal questions without sharing raw data. It organizes the literature around three decisions that any solution must make: the approach to learning structures, the way data are split across parties, and the structural information visible to each party. These decisions produce three taxonomies covering methodological choices, federation layouts, and the scope of shared structure. The paper also treats causal discovery and inference as linked steps in one pipeline and reviews practical complications such as heterogeneity and missing values. Readers benefit because the organization reduces the cost of entering a scattered interdisciplinary literature.

Core claim

The paper establishes that any federated causal discovery solution rests on three design decisions—how structures are learned, how data are partitioned, and what structural knowledge each party obtains—and that these decisions define the axes of methodological paradigm, federation topology, and structural scope. It further shows that federated causal inference methods can be grouped by target estimand and by estimation strategy, and that discovery supplies the structure needed for valid inference, forming stages of a single pipeline.

What carries the argument

The three core design decisions (structure learning method, data partitioning scheme, and per-party structural knowledge) that generate the taxonomies of methodological paradigm, federation topology, and structural scope.

If this is right

  • The taxonomies let researchers locate gaps where no methods exist for particular combinations of learning approach, topology, and scope.
  • Treating discovery and inference as sequential stages means that errors in structure recovery directly limit the validity of downstream effect estimates.
  • Practical factors such as temporal dynamics, data heterogeneity, and non-identical variable sets must be handled inside the same three-axis framework.
  • Privacy, communication cost, and theoretical guarantees are shared constraints that apply across both discovery and inference stages.
  • Open problems remain in extending the framework to new application domains and in proving finite-sample guarantees under federation constraints.

Where Pith is reading between the lines

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

  • The same three decisions could be used to classify methods in other privacy-sensitive graphical modeling tasks beyond causality.
  • New methods might deliberately combine choices from different axes to produce hybrid algorithms that trade off privacy and accuracy in controlled ways.
  • Empirical studies could test whether methods placed in the same taxonomy cell behave similarly on benchmark graphs with controlled federation constraints.
  • The pipeline view suggests that joint optimization of discovery and inference stages might reduce error propagation compared with separate treatment.

Load-bearing premise

The three design decisions form a complete and non-overlapping classification that covers every federated causal discovery method.

What would settle it

Publication of an FCD method whose design choices cannot be placed on the three axes without forcing an arbitrary fourth category or creating unavoidable overlap between existing categories.

read the original abstract

Causal reasoning, which encompasses the discovery of causal structures and the inference of causal effects, is fundamental to data-driven decision making. In practice, data for reliable causal analysis are often distributed across institutions and cannot be centralized due to privacy regulations or communication constraints. Federated learning (FL) addresses this by enabling collaborative analysis without raw data sharing, giving rise to the rapidly growing field of federated causal discovery (FCD) and inference (FCI). However, the interdisciplinary nature of this field and the absence of a comprehensive survey present barriers to entry for researchers. This paper bridges that gap by providing a systematic review through multi-dimensional taxonomies. Grounded in the three core design decisions underlying any FCD solution, namely how structures are learned, how data are partitioned, and what structural knowledge each party obtains, we organize FCD along three axes: methodological paradigm, federation topology, and structural scope. We further examine key practical dimensions, including temporal dynamics, data heterogeneity, missing data, and non-identical variable sets. For FCI, we categorize methods by target estimand (average versus individualized/conditional treatment effects) and by estimation strategy, from classical weighting methods to modern deep generative architectures. Unlike prior works that treat FCD and FCI separately, we formalize their connection as complementary stages of a unified federated causal reasoning pipeline, where FCD supplies the structural knowledge required for valid effect estimation in FCI. Finally, we highlight their shared concerns regarding privacy, communication efficiency, theoretical guarantees, and application domains, and conclude by identifying open challenges for future research.

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

1 major / 2 minor

Summary. The manuscript is a survey on federated causal discovery (FCD) and federated causal inference (FCI). It organizes FCD methods via a three-axis taxonomy (methodological paradigm, federation topology, structural scope) derived from three core design decisions (how structures are learned, how data are partitioned, and what structural knowledge each party obtains). It further reviews practical dimensions such as temporal dynamics and heterogeneity, formalizes FCD and FCI as complementary stages of a unified pipeline, categorizes FCI methods by estimand and strategy, and discusses shared concerns including privacy and open challenges.

Significance. If the proposed taxonomy proves comprehensive, the survey would provide a useful entry point and organizational framework for an emerging interdisciplinary area, with the explicit linkage of discovery to inference as a constructive contribution. The coverage of both methodological and practical aspects strengthens its potential utility for researchers.

major comments (1)
  1. [Abstract and taxonomy introduction section] Abstract and the section introducing the taxonomy: the assertion that the three core design decisions constitute a complete, non-overlapping basis for all FCD solutions is presented without an explicit exhaustive mapping of the cited literature onto the resulting 3D grid or discussion of potential overlaps/collisions with the supplementary dimensions (temporal dynamics, heterogeneity, etc.). This leaves the claimed generality of the taxonomy unverified and load-bearing for the central organizational contribution.
minor comments (2)
  1. [Taxonomy figures] The taxonomy diagrams would benefit from explicit cell-by-cell examples drawn from the surveyed papers to improve immediate readability.
  2. Ensure the reference list includes all works mentioned in the text and consider adding a table summarizing the mapping of key papers to the three axes.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive comment on the taxonomy presentation. We address it point by point below.

read point-by-point responses

  1. Referee: [Abstract and taxonomy introduction section] Abstract and the section introducing the taxonomy: the assertion that the three core design decisions constitute a complete, non-overlapping basis for all FCD solutions is presented without an explicit exhaustive mapping of the cited literature onto the resulting 3D grid or discussion of potential overlaps/collisions with the supplementary dimensions (temporal dynamics, heterogeneity, etc.). This leaves the claimed generality of the taxonomy unverified and load-bearing for the central organizational contribution.

    Authors: The three core design decisions (how structures are learned, how data are partitioned, and what structural knowledge each party obtains) are presented as the logical foundation for any FCD solution because they directly encode the fundamental constraints of federated settings. The manuscript then organizes the reviewed literature along the resulting three axes in dedicated sections. We acknowledge that the abstract and taxonomy introduction do not contain an explicit exhaustive mapping table of all cited works onto the 3D grid, nor a dedicated discussion of interactions with the supplementary dimensions. In the revised version we will add both: (i) a summary table mapping representative papers to taxonomy cells to make completeness verifiable, and (ii) a short paragraph clarifying that supplementary dimensions such as temporal dynamics and heterogeneity are treated as orthogonal refinements that operate within the core axes rather than creating overlaps or collisions in the primary classification. This addresses the concern while preserving the taxonomy's grounding in the three design decisions. revision: yes

Circularity Check

0 steps flagged

No circularity: survey aggregates external results without self-referential derivation

full rationale

This is a literature survey paper whose central contribution is a multi-dimensional taxonomy for organizing existing FCD/FCI methods. The three core design decisions (structure learning, data partitioning, structural knowledge per party) are presented as an organizing lens drawn from the literature rather than derived from any fitted parameters, self-citations, or equations within the paper itself. No equations, predictions, or uniqueness theorems are claimed; the work explicitly states it reviews and categorizes published methods. The taxonomy is therefore not equivalent to its inputs by construction, and the paper remains self-contained against external benchmarks. No load-bearing steps reduce to self-definition or fitted inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

As a survey the contribution is taxonomic organization rather than derivation from axioms or parameters; no free parameters, axioms, or invented entities are introduced to support a central claim.

pith-pipeline@v0.9.1-grok · 5828 in / 1126 out tokens · 43687 ms · 2026-06-26T10:21:55.593535+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

29 extracted references · 5 canonical work pages · 1 internal anchor

  1. [1]

    Causation, Prediction, and Search

    1 Spirtes P, Glymour C N, Scheines R. Causation, Prediction, and Search. MIT Press, 2000 2 Chickering D M. Optimal structure identification with greedy search. Journal of Machine Learning Research, 2002, 3: 507–554 3 Zheng X, Aragam B, Ravikumar P K, et al. DAGs with NO TEARS: Continuous optimization for structure learning. Advances in Neural Information ...

    2000

  2. [2]

    Advances and open problems in federated learning

    1273–1282 7 Kairouz P, McMahan H B. Advances and open problems in federated learning. Foundations and Trends in Machine Learning, 2021, 14: 1–210 8 Wright R, Yang Z. Privacy-preserving Bayesian network structure computation on distributed heterogeneous data. In: Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining,

    2021

  3. [3]

    Learning Bayesian network structure from distributed homogeneous data

    713–718 9 Gou K X, Jun G X, Zhao Z. Learning Bayesian network structure from distributed homogeneous data. In: Proceedings of the Acis International Conference on Software Engineering, Artificial Intelligence, Networking, and Parallel/Distributed Computing (SNPD 2007). IEEE,

    2007

  4. [4]

    Federated causal discovery in medicine: Trends, opportunities, and challenges

    250–254 10 Rocchi N, Scutari M, Zanga A, et al. Federated causal discovery in medicine: Trends, opportunities, and challenges. working paper or preprint, November 2025 11 Li Q, Wen Z, Wu Z, et al. A survey on federated learning systems: Vision, hype and reality for data privacy and protection. IEEE Transactions on Knowledge and Data Engineering, 2021, 35:...

    2025

  5. [5]

    Learning high-dimensional directed acyclic graphs with latent and selection variables

    178–184 17 Colombo D, Maathuis M H, Kalisch M, et al. Learning high-dimensional directed acyclic graphs with latent and selection variables. The Annals of Statistics, 2012, 40: 294–321 18 Zhang J. On the completeness of orientation rules for causal discovery in the presence of latent confounders and selection bias. Artificial Intelligence, 2008, 172: 1873...

    2012

  6. [6]

    Gradient-based neural DAG learning

    3414–3425 23 Lachapelle S, Brouillard P, Deleu T, et al. Gradient-based neural DAG learning. In: Proceedings of the International Conference on Learning Representations, 2020 24 Pamfil R, Sriwattanaworachai N, Desai S, et al. DYNOTEARS: Structure learning from time-series data. In: Proceedings of the International Conference on Artificial Intelligence and...

    2020

  7. [7]

    Neural Granger causality

    1595–1605 25 Tank A, Covert I, Foti N, et al. Neural Granger causality. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 44: 4267–4279 26 Tsamardinos I, Brown L E, Aliferis C F. The max-min hill-climbing Bayesian network structure learning algorithm. Machine Learning, 2006, 65: 31–78 27 Nandy P, Hauser A, Maathuis M H. High-dimensiona...

    2021

  8. [8]

    A generalization of sampling without replacement from a finite universe

    Statistical Science, 1990, pages 465–472 29 Horvitz D G, Thompson D J. A generalization of sampling without replacement from a finite universe. Journal of the American Statistical Association, 1952, 47: 663–685 30 Robins J. A new approach to causal inference in mortality studies with a sustained exposure period—application to control of the healthy worker...

    1990

  9. [9]

    Tackling the objective inconsistency problem in heterogeneous federated optimization

    5132–5143 37 Wang J, Liu Q, Liang H, et al. Tackling the objective inconsistency problem in heterogeneous federated optimization. Advances in Neural Information Processing Systems, 2020, 33: 7611–7623 38 Yang Q, Liu Y, Chen T, et al. Federated machine learning: Concept and applications. ACM Transactions on Intelligent Systems and Technology, 2019, 10: 1–1...

    2020

  10. [10]

    Communication-efficient federated learning via knowledge distillation

    22802–22838 43 Wu C, Wu F, Lyu L, et al. Communication-efficient federated learning via knowledge distillation. Nature Communications, 2022, 13: 2032 44 Dwork C, Roth A. The algorithmic foundations of differential privacy. Foundations and Trends®in Theoretical Computer Science, 2014, 9: 211–487 45 Abadi M, Chu A, Goodfellow I, et al. Deep learning with di...

    2022

  11. [11]

    Crypten: Secure multi-party computation meets machine learning

    308–318 46 Knott B, Venkataraman S, Hannun A, et al. Crypten: Secure multi-party computation meets machine learning. Advances in Neural Sci China Inf Sci25 Information Processing Systems, 2021, 34: 4961–4973 47 Bonawitz K, Ivanov V, Kreuter B, et al. Practical secure aggregation for privacy-preserving machine learning. In: Proceedings of the ACM SIGSAC Co...

    2021

  12. [12]

    A fully homomorphic encryption scheme

    1175–1191 48 Gentry C. A fully homomorphic encryption scheme. Stanford university, 2009 49 Acar A, Aksu H, Uluagac A S, et al. A survey on homomorphic encryption schemes: Theory and implementation. ACM Computing Surveys, 2018, 51: 1–35 50 Huang J, Guo X, Yu K, et al. Towards privacy-aware causal structure learning in federated setting. IEEE Transactions o...

    2009

  13. [13]

    Federated causal discovery from heterogeneous data

    351–367 52 Li L, Ng I, Luo G, et al. Federated causal discovery from heterogeneous data. In: Proceedings of the International Conference on Learning Representations, 2024 53 Hahn M, Zajak A, Heider D, et al. Federated causal discovery across heterogeneous datasets under latent confounding. arXiv preprint arXiv:2603.05149, 2026 54 Guo X, Yu K, Liu L, et al...

    work page arXiv 2024

  14. [14]

    Distributed Bayesian network structure learning

    1998–2003 56 Na Y, Yang J. Distributed Bayesian network structure learning. In: Proceedings of the IEEE International Symposium on Industrial Electronics. IEEE,

    1998

  15. [15]

    Federated learning of generalized linear causal networks

    9–16 58 Ye Q, Amini A A, Zhou Q. Federated learning of generalized linear causal networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024, 46: 6623–6636 59 Chen R, Sivakumar K, Kargupta H. Learning Bayesian network structure from distributed data. In: Proceedings of the SIAM International Conference on Data Mining. SIAM,

    2024

  16. [16]

    FedDAG: Federated DAG structure learning

    61–69 62 Gao E, Chen J, Shen L, et al. FedDAG: Federated DAG structure learning. Transactions on Machine Learning Research, 2023 63 Mian O, Kaltenpoth D, Kamp M, et al. Nothing but regrets—privacy-preserving federated causal discovery. In: Proceedings of the International Conference on Artificial Intelligence and Statistics. PMLR,

    2023

  17. [17]

    Interventional causal structure discovery over graphical models with convergence and optimality guarantees

    100–109 68 Qiu C, Yang K. Interventional causal structure discovery over graphical models with convergence and optimality guarantees. IEEE Transactions on Network Science and Engineering, 2024, 12: 156–172 69 Van Daalen F, Ippel L, Dekker A, et al. VertiBayes: learning Bayesian network parameters from vertically partitioned data with missing values. Compl...

    2024

  18. [18]

    Federated local causal structure learning

    164–179 71 Yu K, Rong C, Wang H, et al. Federated local causal structure learning. Science China Information Sciences, 2025, 68: 132105 72 Rong C, Cao D, Yu K. Efficient local causal structure learning with privacy preservation. In: Proceedings of the International Joint Conference on Rough Sets. Springer,

    2025

  19. [19]

    Federated causal structure learning with missing data

    378–392 73 Shi J, Huang X, Guo X, et al. Federated causal structure learning with missing data. Knowledge-Based Systems, 2025, 330: 114601 74 Chen J, Ma Y, Yue X. Federated learning of dynamic Bayesian network via continuous optimization from time series data. IEEE Transactions on Artificial Intelligence, 2025 75 Mohanty A, Mohamed N, Ramanan P, et al. Fe...

    2025

  20. [20]

    Federated multi-task Bayesian network learning in the presence of overlapping and distinct variables

    62445–62466 78 Yang X, Niu B, Lan T, et al. Federated multi-task Bayesian network learning in the presence of overlapping and distinct variables. IISE Transactions, 2025, 57: 773–787 79 Zanga A, Bernasconi A, Lucas P J, et al. Federated causal discovery with missing data in a multicentric study on endometrial cancer. Journal of Biomedical Informatics, 202...

    2025

  21. [21]

    Towards Uncertainty-Aware Federated Granger Causal Learning

    20280–20288 83 Fehri G F E, Bellet A, Bastien P. Differentially private and federated structure learning in Bayesian networks. In: Proceedings of the International Conference on Artificial Intelligence and Statistics, 2026 84 Torrijos P, G´ amez J A, Puerta J M. FedGES: A federated learning approach for Bayesian network structure learning. Machine Learnin...

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  22. [22]

    An adaptive kernel approach to federated learning of heterogeneous causal effects

    2024–2034 Sci China Inf Sci26 89 Vo T V, Wei P, Gkoulalas-Divanis A, et al. An adaptive kernel approach to federated learning of heterogeneous causal effects. In: Proceedings of the Advances in Neural Information Processing Systems,

    2024

  23. [23]

    Federated causal inference in heterogeneous observational data

    25540–25553 90 Xiong R, Koenecke A, Powell M, et al. Federated causal inference in heterogeneous observational data. Statistics in Medicine, 2023, 42: 4418–4439 91 Han L, Shen Z, Zubizarreta J. Multiply robust federated estimation of targeted average treatment effects. Advances in Neural Information Processing Systems, 2023, 36: 70453–70482 92 Almod´ ovar...

    2023

  24. [24]

    Collaborative inference for treatment effect with distributed data-sharing management in multicenter studies

    16849–16868 95 Hu M, Shi X, Song P X K. Collaborative inference for treatment effect with distributed data-sharing management in multicenter studies. Statistics in Medicine, 2024, 43: 2263–2279 96 Kawamata Y, Motai R, Okada Y, et al. Collaborative causal inference on distributed data. Expert Systems with Applications, 2024, 244: 123024 97 Makhija D, Ghosh...

    work page arXiv 2024

  25. [25]

    Heterogeneity-aware federated causal inference leveraging effect-measure transportability

    472–489 114 Cao S, Yang S. Heterogeneity-aware federated causal inference leveraging effect-measure transportability. arXiv preprint arXiv:2510.16317, 2025 115 Almod´ ovar A, Parras J, Zazo S. Propensity weighted federated learning for treatment effect estimation in distributed imbalanced envi- ronments. Computers in Biology and Medicine, 2024, 178: 10877...

    work page arXiv 2025

  26. [26]

    Toward causal representation learning

    11–20 121 Sch¨ olkopf B, Locatello F, Bauer S, et al. Toward causal representation learning. Proceedings of the IEEE, 2021, 109: 612–634 122 Athey S, Imbens G W. The state of applied econometrics: Causality and policy evaluation. Journal of Economic Perspectives, 2017, 31: 3–32 123 Chernozhukov V, Chetverikov D, Demirer M, et al. Double/debiased machine l...

    2021

  27. [27]

    Dense: Data-free one-shot federated learning

    560–569 130 Zhang J, Chen C, Li B, et al. Dense: Data-free one-shot federated learning. Advances in Neural Information Processing Systems, 2022, 35: 21414–21428 131 Tsiatis A A. Semiparametric theory and missing data. Springer, 2006 132 Murphy S A. Optimal dynamic treatment regimes. Journal of the Royal Statistical Society Series B: Statistical Methodolog...

    2022

  28. [28]

    Causal reasoning and large language models: Opening a new frontier for causality

    11012–11022 Sci China Inf Sci27 135 Kiciman E, Ness R, Sharma A, et al. Causal reasoning and large language models: Opening a new frontier for causality. Transactions on Machine Learning Research, 2024 136 Ban T, Chen L, Wang X, et al. From query tools to causal architects: Harnessing large language models for advanced causal discovery from data. arXiv pr...

    work page arXiv 2024

  29. [29]

    Causality-based feature selection: Methods and evaluations

    673–678 141 Yu K, Guo X, Liu L, et al. Causality-based feature selection: Methods and evaluations. ACM Computing Surveys, 2020, 53: 1–36 142 Gupta V, Luqman A, Chattopadhyay N, et al. TravellingFL: Communication efficient peer-to-peer federated learning. IEEE Transactions on Vehicular Technology, 2023, 73: 5005–5019

    2020