Incorporating Expert Knowledge into Bayesian Causal Discovery of Mixtures of Directed Acyclic Graphs
Researchers have developed a new method for Bayesian causal discovery that can incorporate expert knowledge in heterogeneous domains. This approach extends previous work by allowing for mixtures of causal Bayesian networks, rather than assuming a single causal graph. The proposed variational mixture structure learning method successfully infers these mixtures and improves structure learning performance when informed by expert feedback, as demonstrated on synthetic data and a breast cancer database. AI
IMPACT Introduces a novel approach for incorporating expert knowledge into causal discovery for complex, heterogeneous datasets.