A0535
Title: Causal structure learning with unknown interventions
Authors: Armeen Taeb - ETH Zurich (Switzerland) [presenting]
Abstract: With observational data alone, causal inference is a challenging problem. The task becomes easier when having access to data collected from perturbations of the underlying system, even when the nature of these is unknown. We will describe a body of work that makes algorithmic and theoretical advances for identifying plausible causal mechanisms from such perturbation data. Specifically, in the context of Gaussian linear structural equation models, we first characterize the interventional equivalence class of DAGs. We then leverage these results to study high-dimensional consistency guarantees of a thresholded $l_0$ penalized maximum likelihood estimator for learning said class. Finally, we describe extensions to settings with latent variables.
