Algebraic Bayesian Networks: Empirical Estimates of the Sensitivity of Local Posteriori Inference

Abstract

The article is aimed at studying empirical estimates of the sensitivity of the second task of a posteriori inference in a knowledge pattern. The article presents the results of an experiment on finding the relationship between the distortion of incoming information and the results of learning a knowledge pattern. Formally, information distortions were achieved by changing the estimates of the fixed evidence and finding the norm of the difference between the vectors of the original and the resulting evidence. Obtaining empirical estimates is the first example of studying the second task of a posteriori inference in a knowledge pattern. The relevance of the study is emphasized by the growing popularity of machine learning and, most importantly, data preparation, since ABS is one of the few models that can work with inaccurate data.