Computing intervals of probabilities with simulated annealing and probability trees

Abstract

The paper shows a method to compute a posteriori intervals of probabilities when the initial conditional information is given also with intervals of probabilities. The right way of doing an exact computation is with the associated convex set of probabilities. Probability trees are used to represent these initial conditional convex sets, because they save enormously the required space. This paper proposes a simulated annealing algorithm, using probability trees as the representation of the convex sets, in order to compute the a posteriori intervals.