Causal Preference Elicitation

We propose causal preference elicitation, a Bayesian framework for expert-in-the-loop causal discovery that actively queries local edge relations to concentrate a posterior over directed acyclic graphs (DAGs). From any black-box observational posterior, we model noisy expert judgments with a three-way likelihood over edge existence and direction. Posterior inference uses a flexible particle approximation, and queries are selected by an efficient expected information gain criterion on the expert’s categorical response. Experiments on synthetic graphs, protein signaling data, and a human gene perturbation benchmark show faster posterior concentration and improved recovery of directed effects under tight query budgets.

Many scientific questions involve understanding which factors influence others, such as how genes affect disease or how different biological processes interact. Researchers often use data to build possible explanations, but data alone is often insufficient to determine the correct relationships. Domain experts may have valuable knowledge that is difficult to incorporate systematically into these analyses. This paper introduces a method that combines data-driven analysis with targeted expert feedback. Instead of asking experts to specify an entire network of relationships, the method asks a sequence of simple questions about whether one factor directly influences another. Each answer is used to update the system’s current understanding and to choose the next most informative question. Experiments on simulated and real biological datasets show that this approach can identify correct relationships more quickly and accurately than alternative querying strategies. The results suggest that carefully designed expert feedback can substantially improve scientific discovery when data alone leaves important uncertainties unresolved.