Reproduce “Structure Learning on Clustered Data” 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
Ryan Thompson, Matt P. Wand, Veerabhadran Baladandayuthapani
Researchers from the University of Technology Sydney and the University of Michigan developed a "mixed DAG" framework for learning causal structures in clustered data, decomposing cluster-specific effects into population-level fixed effects and individual random effects. This approach ensures acyclicity of the combined graph through a novel differentiable constraint and demonstrates superior performance in structural recovery and reconstruction error compared to methods assuming data homogeneity.
Key insights
- Decomposing causal effects into fixed (population-level) and random (cluster-specific) components effectively models heterogeneity in clustered data.
- A differentiable formulation of the acyclicity constraint, applied to the union of fixed and random effects graphs, enables scalable gradient-based optimization while ensuring valid causal structures.
- The proposed mixed DAG framework offers theoretical guarantees, including identifiability of parameters and asymptotic consistency for both parameter estimation and structural recovery.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.