Greedy Learning of Causal Structures in Additive Noise Models
Published in Journal 1, 2019
In my thesis, I proved that a Greedy-algorithm can recover the true causal structure of a certain class of structural causal models when supplied with the distribution of this causal model. This algorithm was first proposed by my advisor, Jonas Peters (UCPH), and Martin Wainwright. My main task in this paper was to prove that the algorithm does indeed recover the true causal structure. Furthermore, I was able to extend the class of SCM’s on which this algorithm works from a class of polytrees to a larger class, which is characterized by all undirected cycles having at least three colliders. We expect to rewrite this as an article at a later point in time.