|
Abstract : |
Temporal formalisms are useful in several applications such as planning, scheduling and diagnosis. Probabilistic temporal reasoning emerged to deal with the uncertainties usually encountered in such applications. Bayesian networks provide a simple compact graphical representation of a probability distribution by exploiting conditional independencies. This paper presents a simple technique for representing time in Bayesian networks by expressing probabilities as functions of time. Probability transfer functions allow the formalism to deal with causal relations and dependencies between time points. Techniques to represent related time instants are distinct from those used to represent independent time instants but the probabilistic formalism is useful in both cases. The study of the cumulative effect of repeated events involves various models such as the competing risks model and the additive model. Dynamic Bayesian networks inference mechanisms are adequate for temporal probabilistic reasoning described in this work. Examples from medical diagnosis, circuit diagnosis and common sense reasoning help illustrate the use of these techniques., |
