B1622
Title: Structural restrictions in local causal discovery: Identifying direct causes of a target variable
Authors: Juraj Bodik - UNIL Lausanne (Switzerland) [presenting]
Valerie Chavez-Demoulin - University of Lausanne (Switzerland)
Abstract: The problem of learning a set of direct causes of a target variable from an observational joint distribution is considered. Several results are known when the directed acyclic graph (DAG) is identifiable from the distribution, such as assuming a nonlinear Gaussian data-generating process. Often, the only interest is identifying the direct causes of one target variable (local causal structure), not the full DAG. Different assumptions for the data-generating process of the target variable are discussed under which the set of direct causes is identifiable from the distribution. While doing so, no assumptions are put on the variables other than the target variable. In addition to the novel identifiability results, two practical algorithms are provided for estimating the direct causes from a finite random sample and demonstrate their effectiveness on several benchmark datasets. The framework is applied to learn the direct causes of the reduction in fertility rates in different countries.
