Reproduce “Federated Causal Discovery Across Heterogeneous Datasets under Latent Confounding” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Maximilian Hahn, Alina Zajak, Dominik Heider, Adèle Helena Ribeiro
fedCI-IOD combines federated conditional-independence testing with the IOD algorithm to discover causal structure across sites without moving raw observations. fedCI fits shared generalized linear models from masked, aggregated contributions, including when sites measure different variables. In synthetic tests with 500 samples and 12 partitions, its best normalized SHD was 0.341, close to pooled IOD’s 0.344 and below Fisher meta-analysis’s 0.403; this compares PAG lists with a pooled-oracle reference, not a known true graph.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.