Speaker: Mathias DrtonLink opens in a new windoLink opens in a new window, Technical University of Munich

Title: Causal discovery with graphical models

Abstract: Causal discovery is the problem of inferring cause-effect relationships among a set of variables on the basis of multivariate data, where these variables are jointly observed. Common methods to tackle this problem are based on directed graphical models, which are able to tractably capture stochastic dependencies resulting from the causal relations. In this framework, methods for causal discovery solve the model selection problem of inferring the graph underlying the directed graphical model. In the talk, I will review key ideas in causal discovery and present recent projects on learning from conditional independences and from special distributional assumptions in the form of linear non-Gaussian models.