Causal inference and causal discovery
Distinguishing causation from correlation is one of the most challenging problems in the analysis of any kind of data. Causal inference is the problem of estimating the magnitude of a causal effect if a qualitative graph of causal relations, including unmeasured confounding variables, is known. Causal discovery by contrast is the problem of learning such a causal graph directly from data when no prior knowledge is available.
I am interested in both problems, particularly where they begin to intersect. In most application domains we have some generic knowledge of how information flows between certain classes of events or variables, even if we don’t know the actual causal graph between them. The question I am interested in is how this type of knowledge can help with the discovery of causal interactions?
For instance in biology, the central dogma or the laws genetic inheritance provide generic principles that can be exploited in biological network inference.