A. V. Sirotkin

The paper describes several different models for processing inconsistent data. In case of local inference, there are two principally different models. The first model traditionally cuts all possibilities that contradict consistency conditions, and such way the estimates are narrowing (growing more exact). In the case of inconsistent data, such models entail that there is no way to make a decision. On the other hand, the second model offers to concern initial data by enlarging estimation intervals, such way that obtained estimation is minimal that make knowledge pattern consistent. For these two models the paper presents computational complexity estimates. In the case of global inference, there are four data models under consideration. These models are named global consistency, internal consistency, external consistency, and local consistency. These models have different strength. There is an order for those models such that if algebraic Bayesian networks are consistent in terms of stronger models then these ABNs are consistent in term of weaker models. The paper presents computational complexity estimates system that shows that stronger models have harder algorithms.