A0313
Title: Confidence sets for causal discovery
Authors: Mladen Kolar - University of Chicago (United States) [presenting]
Abstract: Causal discovery procedures are popular methods for discovering causal structures across the physical, biological, and social sciences. However, most procedures for causal discovery only output a single estimated causal model or single equivalence class of models. A procedure is proposed for quantifying uncertainty in causal discovery. Specifically, linear structural equation models are considered with non-Gaussian errors and propose a procedure that returns a confidence set of causal orderings not ruled out by the data. It is shown that asymptotically, the true causal ordering will be contained in the returned set with some user-specified probability.
