Wangyx_2021-P-16224_MMD-LCS_datanet

We consider the problem of distinguishing direct causes from direct effects of a targetvariable from multiple manipulated datasets with unknown manipulated variables and nonidenticaldata distributions. Recent studies have shown that datasets attained from manipulated experiments(i.e. manipulated data) contain richer causal information than observational data for causalstructure learning. Thus in this paper, we propose a new local causal structure learning algorithmwhich makes full use of the interventional properties of a causal model to find the parents andchildren of a target variable from multiple datasets with different manipulations. It is more suitedto real-world cases, and is also a challenge to be addressed in this paper. First, we apply thebackward framework to learn PC (parents and children) of a given target from multiplemanipulated datasets. Second, we orient some edges connected to the target in advance throughthe assumption that the target is not manipulated, and then orient the remaining undirected edgesby finding invariant V-structures from multiple datasets. Third, we analyze the correctness of ourproposed algorithm. To the best of our knowledge, the proposed algorithm is the first that canidentify the local causal structure of a given target from multiple manipulated datasets withunknown manipulated variables. Experimental results on standard Bayesian networks show thatour algorithm outperforms existing algorithms.