Research map / Federated learning and privacy / Personalized and heterogeneous FL
Federated Causal Discovery from Heterogeneous Data
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes FedCDH, a constraint-based federated causal discovery method for heterogeneous data using a surrogate client-index variable, a federated conditional independence test and a federated independent change principle, both computed from summary statistics (sample size and random-feature covariance tensor).
Problem
Federated causal discovery is limited by assumptions of identifiable functional causal models and homogeneous data distributions across clients; the paper targets decentralized, heterogeneous data with arbitrary causal models.
Approach
Uses the client/domain index as a surrogate variable to model heterogeneity, and makes conditional independence tests and the independent change principle federated by computing them from summed local covariance tensors built with random Fourier features, so raw data are never shared and only one communication round is needed.
Key results
- On synthetic data FedCDH generally outperforms FedPC, FedDAG, NOTEARS-ADMM and NOTEARS-MLP-ADMM across varying numbers of variables, clients and samples, averaged over 10 seeds.
- Summary statistics require only one communication round between clients and server, versus iterative communication in baselines.
- Performance improves with more samples per client, consistent with the asymptotic theory.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use FedCDH instead of the compared FCD baselines | Federated causal discovery · Synthetic linear Gaussian and general functional models | F1 score | Helped |
| Use FedCDH instead of the compared FCD baselines | Federated causal discovery · Synthetic linear Gaussian and general functional models | SHD | Helped |
Limitations
- Communication savings diminish when sample size is small or random feature dimension is large.
- Only applicable to horizontally partitioned data, not vertically partitioned data.
- Gaussian kernel assumes smoothness for continuous variables; secure computation not implemented.
Experimental setup
Datasets: fMRI Hippocampus HK Stock Market synthetic linear Gaussian and general functional SCM data
Benchmarks: fMRI Hippocampus HK Stock Market synthetic linear Gaussian and general functional datasets
Metrics: skeleton F1 skeleton SHD direction F1 direction SHD precision recall
Baselines: NOTEARS-ADMM NOTEARS-MLP-ADMM FedDAG FedPC
Builds on: CD-NOD KCI Independent Change Principle PC algorithm random Fourier features
Closest papers by meaning
- Federated Causality Learning with Explainable Adaptive Optimization (AAAI 2024 · similarity 0.90)
- Federated Causal Structure Learning with Non-identical Variable Sets (ICML 2025 · similarity 0.88)
- On Causal Discovery in the Presence of Deterministic Relations (NeurIPS 2024 · similarity 0.87)
- FedCSL: A Scalable and Accurate Approach to Federated Causal Structure Learning (AAAI 2024 · similarity 0.87)
- On the Recoverability of Causal Relations from Temporally Aggregated I.I.D. Data (ICML 2024 · similarity 0.87)
- Causal Discovery via Conditional Independence Testing with Proxy Variables (ICML 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.