Towards a Learning Theory of Cause-Effect Inference

We pose causal inference as the problem of learning to classify probability distributions. In particular, we assume access to a collection {(Si,li)}ni=1, where each Si is a sample drawn from the proba