engleski

Object oriented graphical environment for numerical approach to uncertainty management

The problem of imperfect information in intelligent systems is handled in two distinctive ways: 1) symbolically, in the case of incomplete information, and 2) numerically, in the case of imprecise and uncertain information. Symbolic methods are based upon extensions to classical logic, while numerical methods employ quantifiers that may propagate over a network representing dependencies between relevant facts. The paper will concentrate on the network models to represent uncertainty, following the idea of adjusting some underlying probability distribution which makes particular options the correct conclusions. One of the problems with network models is their inherent complex abstraction. The paper will analyze all requirements for a visual probabilistic dependency network design environment, suitable for numerical uncertainty management. A specific object oriented framework will be revealed and discussed.

probabilistic reasoning; artificial intelligence; Bayesian networks

nije evidentirano