2013/02/27 by Eugene Santos, Santos, Eugene, Solomon Eyal Shimony +2
Computer Science · Decision Sciences · #Artificial Intelligence (cs.AI) #Bayesian Modeling and Causal Inference #Explainable Artificial Intelligence (XAI) #FOS: Computer and information sciences #Multi-Criteria Decision Making
paper · pdf · doi:10.48550/arxiv.1302.6842
openalex publication_date 2013/02/27 · openalex created_date 2022/10/05 · openalex updated_date 2026/07/28
Independence-based (IB) assignments to Bayesian belief networks were\noriginally proposed as abductive explanations. IB assignments assign fewer\nvariables in abductive explanations than do schemes assigning values to all\nevidentially supported variables. We use IB assignments to approximate marginal\nprobabilities in Bayesian belief networks. Recent work in belief updating for\nBayes networks attempts to approximate posterior probabilities by finding a\nsmall number of the highest probability complete (or perhaps evidentially\nsupported) assignments. Under certain assumptions, the probability mass in the\nunion of these assignments is sufficient to obtain a good approximation. Such\nmethods are especially useful for highly-connected networks, where the maximum\nclique size or the cutset size make the standard algorithms intractable. Since\nIB assignments contain fewer assigned variables, the probability mass in each\nassignment is greater than in the respective complete assignment. Thus, fewer\nIB assignments are sufficient, and a good approximation can be obtained more\nefficiently. IB assignments can be used for efficiently approximating posterior\nnode probabilities even in cases which do not obey the rather strict skewness\nassumptions used in previous research. Two algorithms for finding the high\nprobability IB assignments are suggested: one by doing a best-first heuristic\nsearch, and another by special-purpose integer linear programming. Experimental\nresults show that this approach is feasible for highly connected belief\nnetworks.\n