On automatic knowledge validation for Bayesian knowledge bases

Knowledge validation, as part of knowledge base verification and validation, is a critical process in knowledge engineering. The ultimate goal of this process is to make the knowledge base satisfy all test cases given by human experts. This is further complicated by factors such as uncertainty and incompleteness. Our paper covers theoretical results in knowledge validation for Bayesian Knowledge Bases (BKBs), a probabilistic model extended from Bayesian Networks for representing knowledge in uncertain domains. First, we study the consistency of test case sets by identifying the necessary and sufficient conditions for a test case set such that there exists a knowledge base satisfying all of its test cases. Second, we analyze the thrashing problem which is the interminable oscillation of the knowledge base’s state when validating by parameter refinement. We propose an approach to validating BKBs that effectively eliminates thrashing under certain conditions of the original knowledge base and the test case set.