New survey maps federated causal discovery and inference for privacy-preserving analysis

Causal reasoning – discovering causal structures and estimating causal effects – is critical for data-driven decisions, but real-world data for such analysis is often distributed across institutions and cannot be centralized due to privacy regulations like GDPR. Federated learning offers a solution by enabling collaborative analysis without raw data sharing, giving rise to the rapidly growing field of federated causal discovery (FCD) and inference (FCI). In a new 27-page survey paper on arXiv, Xianjie Guo and six other researchers provide the first comprehensive review bridging these two previously separate domains. They ground their analysis in three core design decisions underlying any FCD solution: how structures are learned, how data are partitioned, and what structural knowledge each party obtains. From there, they organize FCD along three axes – methodological paradigm, federation topology, and structural scope – and examine practical dimensions like temporal dynamics, data heterogeneity, missing data, and non-identical variable sets.

For FCI, the survey categorizes methods by target estimand (average treatment effects versus individualized/conditional treatment effects) and by estimation strategy, ranging from classical weighting methods to modern deep generative architectures. A key contribution is formalizing FCD and FCI as complementary stages of a unified federated causal reasoning pipeline: FCD supplies the structural knowledge necessary for valid effect estimation in FCI. The paper also highlights shared concerns around privacy, communication efficiency, theoretical guarantees, and application domains (e.g., healthcare, finance). By providing multi-dimensional taxonomies and identifying open challenges, this survey aims to lower entry barriers for researchers and accelerate progress in privacy-preserving causal analysis across federated networks.