Catalogue of Artificial Intelligence Techniques

   

Jump to: Top | Entry | References | Comments

View Maths as: Images | MathML

Inferno

Keywords: backtracking

Categories: Inference and Reasoning


Author(s): Robert Corlett

Inferno is a conservative approach to uncertainty. It is based on a probabilistic approach to uncertainty, but, unlike most similar schemes, does not make any assumptions about relationships between propositions (e.g., independence). It is, however, possible to assert such relationships in Inferno if they exist. The method is essentially based on upper and lower probabilities (see Dempster-Shafer Theory) which give upper and lower bounds on the probability of each proposition. Each time one of these bounds changes, the bounds on related propositions are checked and suitably modified to satisfy a set of propagation constraints. These constraints are based on the inequality:

max(P(A),P(B))P(AorB)P(A)+P(B)

One unique feature of Inferno is its ability to detect and suggest corrections for contradictions and inconsistencies in its data. If the propagation constraints cannot provide valid values for the bounds, Inferno invokes a method similar to Dependency Directed Backtracking to suggest changes to the data that would remove the inconsistency.


References:


Comments:

Add Comment

No comments.