Federated Causal Inference from Multi-Site Observational Data via Propensity Score Aggregation

Researchers have proposed a Federated Learning approach to estimate the Average Treatment Effect from decentralized observational data, addressing privacy and logistical constraints in causal inference.

The method, detailed in a paper submitted to arXiv^[1], estimates propensity scores via a federated weighted average of local scores using Membership Weights. This allows for the construction of Federated Inverse Propensity Weighting (Fed-IPW) and Augmented IPW (Fed-AIPW) estimators. The approach exploits heterogeneity in treatment assignment across sites to improve overlap, showing clear advantages over meta-analysis in theoretical analysis and experiments on simulated and real-world data. A separate study on arXiv^[2] introduced SDVDiag, a multimodal causal-discovery pipeline for online diagnosis in software-defined vehicles, which fuses log-based and metric-based service representations into a shared embedding space. SDVDiag produces sparser causal graphs than a metrics-only baseline and consistently outperforms it in edge-weighted reward against an expert knowledge graph. The integrated trigger in SDVDiag correctly recovers a true root cause located two causal hops upstream of the observable symptom^[2].