Modularizing inference in large causal probabilistic networks

✍ Scribed by Kristian G. Olesen; Steen Andreassen; Marko Suojanen

• Publisher: John Wiley and Sons
• Year: 2003
• Tongue: English
• Weight: 132 KB
• Volume: 18
• Category: Article
• ISSN: 0884-8173
• DOI: 10.1002/int.10082

✦ Synopsis

This article describes a number of implementation aspects of modular inference in large medical expert systems based on causal probabilistic networks. Examples are provided from the neuromuscular diagnosting system the muscle and nerve inference network (MUNIN). The inference procedure is outlined and the principal data structure underlying the inference procedure are described. A condensed summary of selected technical details of the inference procedure in causal probabilistic networks (CPNs) is provided. This is required for understanding the implemented modularization of the inference. The modularization of the inference implies a need for transfer of information between modules, which is realized by establishing communication channels between modules. Modules are also used to perform inference by conditioning, a method that reduces storage requirements to a manageable size and thereby prepares the way for MUNINs migration to common PCs.