The eclectic approach to modeling uncertainty argues that the best model and associated representation should be used for each application, whereas the unified approach attempts to find a single representation for uncertainty that is sufficiently general to allow it to serve as a common uncertainty interchange format. Specific approaches to modeling and representing uncertainty include both numerical symbolic processing and representation models. Some of the better-known examples of numerical models include probability theory, Bayesian theory, Dempster-Shafer calculus, Fuzzy logic, and certainty factors. Non-motonic logic is an example of a symbolic approach. Relative to the amount of research involving these specific approaches, work to date on unified uncertainty modeling is rather thin. Notable exceptions include the work by Baroni, Guida & Vicig involving uncertainty interchange formats for multi-agent systems.
The heterogeneous, and modular nature of our analysis framework necessitates that various forms of uncertainty representation be supported. Even though individual reasoning technologies may dictate a particular uncertainty model and representation, it is theoretically possible to translate those values to and from a sufficiently expressive uncertainty interchange representation. Although there has been some promising work done in this regard, the current lack of a widely accepted and sufficiently mature unified uncertainty interchange format leads us to believe that the more prudent plan of attack is to support simultaneously both specific and unified representations. Supporting both representations will allow the use of the unified version where it is sufficiently effective while still allowing direct access to the original, un-translated version in special situations.