From adaptive to progressive combination of possibility distributions

Possibilistic information fusion using maximal coherent subsets

“…Associativity is incompatible with adaptiveness and convexity hides the presence of conflict between sources. Overall, the MCS method meets all requirements advocated in [26], [28].…”

“…In [26], alternatives to equation (5) are proposed that take into account the distance between possibility distributions, thus accounting for the metric structure of space X. All these alternatives mainly consist of reformulating π (m) into a distribution π (m) .…”

“…This property is informally stated in [26] as the requirement that all distributions π i should be taken into account (unless explicitly stated otherwise by additional assumptions). Similar properties are more formally stated in [8], where they are called reconciliation and strong reconciliation.…”

“…should only cause a small change on the final result. Since information is often approximately modeled, this property is often considered as desirable [26] (viz. the lack of robustness for the rule given by equation (5), studied in [26]).…”

“…Since information is often approximately modeled, this property is often considered as desirable [26] (viz. the lack of robustness for the rule given by equation (5), studied in [26]). Concerning the method proposed here, although small changes can have an important impact for a particular E α by making a coherent maximal subset no longer coherent, small changes will have small impact on the overall structure (F, m) and on the distribution π c (most of the time, small changes will only cause small shifts in values of β k ).…”