Quite often the state of information provided by disparate sources, or experts, is not conducive to fusion without additional processing. The focus of this paper involves processing information to help improve fusion results in classification problems. Specifically, this paper presents and integrates three necessary techniques to address assumptions about expert information, which are misaligned with reality. These three alignment techniques address the following three causes of misalignment: (1) over confident predictions made by experts, (2) different (but relatable) sets of classifications used by experts to make predictions, and (3) ambiguity in object-to-object association between experts. Addressing these three causes of misalignment avoids the inappropriate weighting, the improper interpretation, and/or the misapplication of expert information, any of which could lead to poor fusion results. Additionally, an expert's use of poorly informed priors (in the Bayesian sense) can prevent the alignment techniques from being most effective, which makes information representation at the credal level (in the transferable belief model sense) an ideal domain to execute these alignment techniques. Zadeh's and another similar example are used to demonstrate the necessity of addressing assumptions misaligned with reality to improve fusion results.
Addressing Assumptions About Expert Information Leveraging the Transferable Belief Model
2021-03-06
15870356 byte
Conference paper
Electronic Resource
English
USING THE TRANSFERABLE BELIEF MODEL TO VEHICLE NAVIGATION SYSTEM
British Library Conference Proceedings | 2006
|