Parallel and Sequential Implementations for Combining Belief Functions
Abstract
This paper reports our experiments about parallel and sequential implementations for combining belief functions with an application to a medical diagnostic system. We use as a basis existing methods for combining two belief functions: a direct combination based on Dempster's rule and an indirect combination through Mobius transforms. We further explore various parallel algorithms for combining more than two belief functions, as different belief functions can be combined in any order as long as they are independent of each other. Our results indicate that for the general case, the parallel implementation based on fast Mobius transforms proves to be the most efficient. However, for practical applications where most subsets of a frame of hypotheses have zero probabilities, the parallel implementation based on an improved direct combination rule remains the most efficient.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1992
- Accession Number
- ADP007126
Entities
People
- Fei Song
- Mary Mcleish
Organizations
- University of Guelph