Improving Causal Discovery By Optimal Bayesian Network Learning

Many widely-used causal discovery methods such as Greedy Equivalent Search (GES), although with asymptotic correctness guarantees, have been reported to produce sub-optimal solutions on finite data, or when the faithfulness condition is violated. The constraint-based procedure Boolean satisfiability (SAT) solver, and recently proposed Sparsest Permutation (SP) algorithm shown superb performance, but currently they do not scale well. In this work, we demonstrate that optimal score-based exhaustive search remarkably useful for discovery: it requires weaker conditions guarantee correctness, outperforms well-known including PC, GES, GSP, NOTEARS. order achieve scalability, also develop an approximation larger systems based A* method, which scales up 60+ variables obtains better results than existing greedy algorithms MMHC, GSP. Our illustrate risk of assuming assumption, advantages methods, limitations shed light computational challenges techniques in scaling networks handling unfaithful data.