In this paper we present an application where we put together two methods for clustering and classification into a force aggregation method. Both methods are based on conflicts between elements. These methods work with different type of elements (intelligence reports, vehicles, military units) on different hierarchical levels using specific conflict assessment methods on each level. We use Dempster-Shafer theory for conflict calculation between elements, Dempster-Shafer clustering for clustering these elements, and templates for classification. The result of these processes is a complete force aggregation on all levels handled.

Abstract- The success of information fusion is defined by the quality of knowledge produced by fusion processes, with the latter in turn depending on how well data are represented, how reliable and adequate the model of data uncertainty used, and how accurate and appropriate or applicable prior knowledge is. The majority of fusion operators is based on optimistic assumptions about reliability of sources and presumes that they are all reliable. At the same time, different sources may have different reliability and it is necessary to account for this fact to avoid decreasing in performance of fusion results. The objective of this paper is to discuss the principal concepts and strategies of incorporating reliability into classical fusion operators and to provide an overview of the main approaches used in the fusion literature.

This paper presents a new classifier combination technique based on the Dempster-Shafer theory of evidence. The Dempster-Shafer theory of evidence is a powerful method for combining measures of evidence from different classifiers. However, since each of the available methods that estimates the evidence of classifiers has its own limitations, we propose here a new implementation which adapts to training data so that the overall mean square error is minimized. The proposed technique is shown to outperform most available classifier combination methods when tested on three different classification problems.

This paper provides a modal logic framework for reasoning about multi-agent belief and its fusion. We propose logics for reasoning about cautiously merged agent beliefs that have different degrees of reliability. These logics are obtained by combining the multi-agent epistemic logic and multisource reasoning systems. The fusion is cautious in the sense that if an agent’s belief is in conflict with those of higher priorities, then his belief is completely discarded from the merged result. We consider two strategies for the cautious merging of beliefs. In the first, called level cutting fusion, if inconsistency occurs at some level, then all beliefs at the lower levels are discarded simultaneously. In the second, called level skipping fusion, only the level at which the inconsistency occurs is skipped. We present the formal semantics and axiomatic systems for these two strategies and discuss some applications of the proposed logical systems. We also develop a tableau proof system for the logics and prove the complexity result for the satisfiability and validity problems of these logics.

In this paper we develop methods for selection of templates and use these templates to recluster an already performed Dempster-Shafer clustering taking into account intelligence to template fit during the reclustering phase. By this process the risk of erroneous force aggregation based on some misplace pieces of evidence from the first clustering process is greatly reduced. Finally, a more reliable force aggregation is performed using the result of the second clustering. These steps are taken in order to maintain most of the excellent computational performance of Dempster-Shafer clustering, while at the same time improve on the clustering result by including some higher relations among intelligence reports described by the templates. The new improved algorithm has a computational complexity of O(n^3 log^2 n) compared to O(n^2 log^2 n) of standard Dempster-Shafer clustering using Potts spin mean field theory.

Abstract. The Transferable Belief Model is a general framework for managing imprecise and uncertain information using belief functions. In this framework, the discounting operation allows to combine information provided by a source (in the form of a belief function) with metaknowledge regarding the reliability of that source, to compute a “weakened”, less informative belief function. In this article, an extension of the discounting operation is proposed, allowing to make use of more detailed information regarding the reliability of the source in different contexts, a context being defined as a subset of the frame of discernment. Some properties of this contextual discounting operation are studied, and its relationship with classical discounted is explained. 1

Abstract. A novel and efficient approach to facial expression classification based on the belief theory and data fusion is presented and discussed. The considered expressions correspond to three (joy, surprise, disgust) of the six universal emotions as well as the neutral expression. A robust contour segmentation technique is used to generate an expression skeleton with facial permanent features (mouth, eyes and eyebrows). This skeleton is used to determine the facial features deformations occurring when an expression is present on the face defining a set of characteristic distances. In order to be able to recognize “pure ” as well as “mixtures ” of facial expressions, a belief-theory based fusion process is proposed. The performances and the limits of the proposed recognition method are highlighted thanks to the analysis of a great number of results on three different test databases: the Hammal-Caplier database, the Cohn-Kanade database and the Cottrel database. Preliminary results demonstrate the interest of the proposed approach, as well as its ability to recognize non separable facial expressions. 1

The paper presents an approach to joint tracking and classification based on belief func-tions as understood in the transferable belief model (TBM). The TBM model is identical to the classical model except all probability functions are replaced by belief functions, which are more flexible for representing uncertainty. It is felt that the tracking phase is well han-dled by the classical Kalman filter but that the classification phase deserves amelioration. For the tracking phase, we derive a minimal set of assumptions needed in the TBM ap-proach in order to recover the classical relations. For the classification phase, we distinguish between the observed target behaviors and the underlying target classes which are usually not in one-to-one correspondence. We feel the results obtained with the TBM approach are more reasonable than those obtained with the corresponding Bayesian classifiers.

In this paper we develop a method for report level tracking based on Dempster-Shafer clustering using Potts spin neural networks where clusters of incoming reports are gradually fused into existing tracks, one cluster for each track. Incoming reports are put into a cluster and continuous reclustering of older reports is made in order to obtain maximum association fit within the cluster and towards the track. Over time, the oldest reports of the cluster leave the cluster for the fixed track at the same rate as new incoming reports are put into it. Fusing reports to existing tracks in this fashion allows us to take account of both existing tracks and the probable future of each track, as represented by younger reports within the corresponding cluster. This gives us a robust report-to-track association. Compared to clustering of all available reports this approach is computationally faster and has a better report-to-track association than simple step-by-step association.

The paper presents an approach to joint tracking and classification based on belief functions as understood in the transferable belief model (TBM). For the tracking phase, a Kalman filter in the TBM framework is derived. This filter is essentially the same as the classical Kalman filter with a di#use prior, although it is derived in a more general context. For the classification phase, the TBM solution provides more reasonable results than the corresponding Bayesian classifier in situations where no one-toone mapping between target behaviours and classes can be established.