首页 >> 科研与学科建设 >> 科学研究 >> 学术报告

Advances and Applications of DSmT for Information Fusion

发布日期:2012-10-31
dot 讲座人:Dr. Florentin Smarandache     dot 讲座时间:2011年12月22日: 上午11点 到12点     dot 讲座地点:理学院会议室(7215)

dot 讲座内容:

The combination of information is a hot topic of research especially in the development of complex systems involving imprecise, uncertain and potentially highly conflicting information/data with usually (but not necessarily) human interaction at some higher fusion level for efficient decision-making. Modern multisensor systems for tracking, classification, diagnosis, situation assessment, etc. need solid theoretical tools to combine efficiently information in order to reduce as best as possible ignorances and contradictions in a coherent way to help to take proper decision. This task is very difficult and many theories probability theory, possibility theory, Dempster-Shafer Theory (DST), etc.) have been proposed to deal with different kinds of uncertainties randomness, fuzziness, epistemic nature, etc.). After a brief reminder of classical combination rules based on belief functions used up to now in most of (non-Bayesian) multisensor/expert systems, a detailed presentation of foundations and advances obtained in the development of Dezert-Smarandache Theory (DSmT) for the combination of uncertain, imprecise, and potentially highly contradicting sources of information will be given. DSmT appears as a natural extension of DST because DSmT takes into consideration any kind of model (free, hybrid DSm models, and alsothe classical Shafer‘s model) according to the integrity constraints of the fusion problem. DSmT proposes a new mathematical framework and rules for information fusion that potentially allows some intersections of elements of the frame (i.e. some degree of consensus between elements). Fusion rules developed in DSmT framework overcome limitations of Dempster‘s rule and its alternatives as it will be showed through very simple examples. DSmT appears well adapted to static or dynamic fusion applications represented in terms of belief functions based on the same unified general mathematical formalism. The mathematical level of this tutorial and of the didactic examples will be kept as simple as possible to show the advantages of this new approach over previous ones. Aside basis of DSmT, we will present the recent Proportional Conflict Redistribution (PCR5) rules and show their performances on several examples and will present also a new general arithmetic for the fusion of qualitative beliefs. An introduction to new quantitative belief conditioning rules will be also presented. A direct extension of the quantitative/numerical information fusion and conditioning rules to their qualitative counterparts in order to deal with qualitative information drawn from human sources and expressed in natural language will be shown. An introduction to Non-Bayesian Reasoning (NBR) and the fusion of sources with different importances will complete this tutorial.

修改日期:2012-2-14