information fusion模板Information FusionOdile PapiniLSIS-CNRS,Université de Méditerranée, ESIL,163 av. de Luminy - 13288 Marseille Cedex 09. Francepapini@esil.univmed.frMerging information coming from d ifferent sources is an important issue in var-ious domains of com puter science like knowledge representation for artificial in-tel ligence, decision making or databases. The aim of fusion is to ob tain a globalpoint of view, exploiting the complementarity betwee n sources, solving differentexisting conflicts, reducing the poss ible redundancies.When focusing on merging, one has to pay attent ion to the nature of thetargeted information to be merged: belief s, generic knowledge, goals or prefer-ences, laws or regulations since the kind of fusion deeply depends on the natureof informati on provided by the sources [I]. Beliefs are factual information, theyrepresent agent''s perceptions or observations and can be fals e. A belief base isan agent''s description of the world according to its perceptions and the fusion ofbelief bases expresses the be liefs of a group of agents on the basis of the invidualbeliefs. O n constrast, generic knowledge is unquestionable information. Whe nmerging generic knowledge coming from different sources, the onl y acceptablefusion method is the conjunction of the information p rovided by the sources andthis conjunction has to be consistent. Moreover, beliefs and generic knowledgerepresent the world as it is assumed to be, however goals or preferences representthe world as it should evolve for the agent and merging goal bases or pref erencebases amounts to find which goals a group of agents should converge to in orderto best satisfy the group. This is related to preference aggregation. Regulations,laws, specifications describ e the world as it should be ideally. The aim of thiskind of fusio n is to provide a consistent base of regulations from several ini tialbases that could conflict.Among the various approaches of mul ti-sources information merging, sym-bolic approaches gave rise to increasing interest within the artificial intelligencecommunity [23456] the last decade. Belief bases merging has received muchat tention and most of the approaches have been defined within the f rameworkof classical logic, more often propositional.Postulates c haracterizing the rational behavior of fusion operations have bee nproposed [7] which capture the following basic assumptions. The sources aremutually independant and no implicit link between the information from thedifferent sources are assumed. All sources ha ve the same level of importance andprovide consistent belief base s. All information from a source have the same levelof reliabilit y or priority. Due to the non-constructive nature of these postul ates,the core problem is the definition of fusion operations. Sev eral merging opera-tions have been proposed that can be divided into two families. The semanticA. Doshpande and A. Hunter (Eds.): SUM 2010, LNAI 6379, pp. 20-23, 2010.? Springer-Verlag Berlin Heidelberg 2010 |
|