|
Abstract : |
To investigate the robustness of its output, a Bayesian network can be subjected to a sensitivity analysis. In such an analysis, basically the relation between the network's output and every single one of its parameters is established. This relation is known to be a simple mathematical function, expressing the network's output as a quotient of two linear functions in a parameter under study. We present a method for computationally efficient sensitivity analysis. Our method requires a single outward propagation in a junction tree for establishing the mathematical functions for the network's output in terms of all its parameters, in addition to a full propagation for processing evidence. Conversely, our method provides for establishing the mathematical functions for all possible outputs in terms of a single parameter under study, also by means of a single outward propagation. 1, |
