Discovering Causal Relations in the Presence of Latent

The causal relationships determining the behaviour of a system under study are inherently directional: by manipulating a cause we can control its ef-fect, but an effect cannot be used to control its cause. Understanding the network of causal relationships is necessary, for example, if we want to predict the behaviour in settings where the system is subject to different manipulations. However, we are rarely able to directly observe the causal processes in action; we only see the statistical associations they induce in the collected data. This thesis considers the discovery of the fundamen-tal causal relationships from data in several different learning settings and under various modeling assumptions. Although the research is mostly theo-retical, possible application areas include biology, medicine, economics and the social sciences. Latent confounders, unobserved common causes of two or more observe