|
Abstract:
|
Learning causal or directional relationships for large-scale multivariate data is an extremely challenging problem. The directed acyclic graphical (DAG) models are a popular class of statistical models for addressing this problem. However, there are a number of statistical and computational challenges associated with learning DAG models from observational data. In this talk I present two of my recent works that investigate and addresses those challenges. Firstly, I introduce a new class of fully identifiable DAG models that apply to a broad class of conditional distributions including Poisson, generalized Poisson, Binomial, Negative Binomial, Gamma, and many others. Secondly, I present a polynomial-time algorithm for learning count-data DAG models based on a computationally feasible approach to learning the causal ordering. My method applies in the setting where the number of variables p is larger than n under degree constraints on the graph.
|
