Reviews: Experimental Design for Learning Causal Graphs with Latent Variables

The authors propose theory and algorithms for identifying ancestral relations, causal edges and latent confounders using hard interventions. Their algorithms assume that it is possible to perform multiple interventions on any set of variables of interest, and the existence of an independence oracle, and thus is mostly of theoretical value. In contrast to previous methods, the proposed algorithms do not assume causal sufficiency, and thus can handle confounded systems. The writing quality of the paper is good, but some parts of it could be changed to improve clarity. General Comments Several parts of the paper are hard to follow (see below).