Reproduce “BaCaDI: Bayesian Causal Discovery with Unknown Interventions” 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
Alexander Hägele, Jonas Rothfuss, Lars Lorch, Vignesh Ram Somnath, Bernhard Schölkopf, Andreas Krause
A Bayesian framework named BaCaDI jointly infers causal structures, mechanism parameters, and unknown intervention targets from multi-context experimental data. This method demonstrated superior accuracy in recovering causal graphs and predicting intervention outcomes compared to existing approaches, particularly under limited data and model misspecification, such as in simulated gene-regulatory networks.
Key insights
- Transforming the inference of discrete causal graphs and intervention targets into a differentiable problem using continuous latent variables allows for efficient, gradient-based variational inference.
- A principled Bayesian treatment, including explicit uncertainty quantification and propagation across causal structure and unknown intervention parameters, enhances robustness and accuracy, particularly with limited data or model misspecification.
- Joint inference of causal structure and unknown intervention targets within a unified framework provides more reliable results than sequential or separate estimation approaches.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.