Computing with Bayesian Multi-Networks
AIR FORCE INST OF TECH WRIGHT-PATTERSON AFB OH SCHOOL OF ENGINEERING
Pagination or Media Count:
Existing probabilistic approaches to automated reasoning impose severe restrictions on its knowledge representation scheme. Mainly, this is to ensure that there exists an effective inferencing algorithm. Unfortunately, this makes the application of these approaches to general domains quite difficult. In this paper, we present a new mode called Bayesian multi-networks which uses a rule-based organization of knowledge quite natural for human experts modeling various domains. Furthermore, strong probabilistic semantics help quantify the knowledge. Combined with the rich structure of rule-based approaches, a general inference engine for Bayesian multi-networks is developed. Probabilistic reasoning, Constraint satisfaction, Linear programming, Temporal reasoning, Abductive explanation.
- Operations Research