Stéphane Loiseau

A Bayesian network based trust model for improving collaboration in mobile ad hoc networks

Functioning as fully decentralised distributed systems, without the need of predefined infrastructures, mobile ad hoc networks provide interesting solutions when setting up dynamic and flexible applications. However, these systems also bring up some problems. In such open environments, it is difficult to discover among the nodes, which are malicious and which are not, in order to be able to choose good partners for cooperation. One solution for this to be possible, is for the entities to be able to evaluate the trust they have in each other and, based on this trust, determine which entities they can cooperate with. In this paper, we present a trust model adapted to ad hoc networks and, more generally, to distributed systems. This model is based on Bayesian networks, a probabilistic tool which provides a flexible means of dealing with probabilistic problems involving causality. The model evaluates the trust in a server according, both, to direct experiences with the server and recommendations concerning its service. We show, through a simulation, that the proposed model can determine the best server out of a set of eligible servers offering a given service. Such a trust model, when applied to ad hoc networks, tends to increase the QoS of the various services used by a host. This, when applied to security related services thus increases the overall security of the hosts.

Malan: a mapping language for the data manipulation

Malan is a MApping LANguage that allows the generation of transformation programs by specifying a schema mapping between a source and target data schema. By working at the schema level, Malan remains independent of any transformation process; it also naturally guarantees the correctness of the transformation target relative to its schema. Moreover, by expressing schemas as UML class diagrams, Malan schema mappings can be written on top of UML modellers. This paper describes the overall approach by focusing on the Malan language itself, and its use within a transformation process.

ONTOLOGICAL COGNITIVE MAP

A cognitive map model provides a graphical representation of an influence network between concepts. One drawback of this model is that large cognitive maps are difficult to exploit and understand. This paper introduces an ontological cognitive map model that enables the designer to organize concepts in an ontology. On the one hand, this model provides an ontological influence mechanism that shows the influence from any concept of the ontology to any other according to the map. The map is then easier to exploit. On the other hand, the ontology is used for providing a synthetical view of a map. The map is then easier to understand.

Validation and Reparation of Knowledge Bases

Two properties characterize the quality of a knowledge base (KB): coherency and completeness. In our work, the completeness of the KB is evaluated on a set of test cases. The incompleteness is revealed by test cases that cannot be proved, called deficiencies. We also use test cases to check the coherency of the KB: we propose a new notion, called C_coherency, that extends previous definitions of coherency. The situations that reveal the C_incoherency of the KB are called conflicts. When the problems in a KB have been identified, our aim is to restore the validity of this base. For each conflict and each deficiency, we determine which modifications of the KB can suppress this problem without introducing new deficiencies or conflicts in the KB.

Ontological Cognitive Map

A cognitive map model provides a graphical representation of an influence network between concepts. One drawback of this model is that large cognitive maps are difficult to exploit and understand.This paper introduces an ontological cognitive map model that enables the designer to organize concepts in an ontology. On one hand, this model provides an ontological influence mechanism that shows the influence from any concept of the ontology to any other according to the map. The map is then easier to exploit. On the other hand, the ontology is used as a scale for providing a synthetical view of a map. The map is then easier to understand.

A Non-parametric Semi-supervised Discretization Method

Semi-supervised classification methods aim to exploit labelled and unlabelled examples to train a predictive model. Most of these approaches make assumptions on the distribution of classes. This article first proposes a new semi-supervised discretization method which adopts very low informative prior on data. This method discretizes the numerical domain of a continuous input variable, while keeping the information relative to the prediction of classes. Then, an in-depth comparison of this semi-supervised method with the original supervised MODL approach is presented. We demonstrate that the semi-supervised approach is asymptotically equivalent to the supervised approach, improved with a post-optimization of the intervals bounds location.

Contextual Cognitive Map

The model of cognitive maps introduced by Tolman [1] provides a representation of an influence network between notions. A cognitive map can contain a lot of influences that makes difficult its exploitation. Moreover these influences are not always relevant for different uses of a map. This paper extends the cognitive map model by describing the validity context of each influence with a conceptual graph. A filtering mechanism of the influences according to a use context is provided so as to obtain a simpler and more adjusted map for a user. A prototype that implements this model of contextual cognitive map has been developed.

Modélisation, classification et propagation dans des réseaux d'influence

The cognitive map model provides a user a solution to visualize the influences between different notion, and to compute the propagation of influences on a target. Like cognitive maps, our model offers a graphical representation of influences between notions. The distinctive feature of our model is that on a unique support, each notion is precisely defined by conceptual graphs. The combination of operations of cognitive maps and operations of conceptual graphs provides a powerful method to make decision. Firstly, the definition of a notion and the projection provides a solution to compute semantically linked notions. Secondly, original propagations can be computed from such semantically linked notions.

Refinement of Conceptual Graphs

The semantic validation of a knowledge base (KB) consists in checking its quality according to constraints given by an expert. The refinement of a KB consists in correcting the errors that are detected during the validation, in order to restore the KB validity. We propose to perform the semantic validation and refinement of a KB composed of conceptual graphs in two stages. First, we study the coherence of the KB with respect to negative constraints, which represent the knowledge that the KB must not contain. When the KB is not coherent, we propose a solution to correct all the errors of the KB. Second, we study the completeness of the KB with respect to positive constraints, which represent the knowledge that the KB must contain. When the KB is not complete, we propose an assistant, which helps the user to correct the errors of the KB one by one.

Synthesis of Cognitive Maps and Applications

Cognitive maps are a knowledge representation model that describes influences between concepts. Their building is usually done by many people. This is a difficult task for them since they have to agree on every aspect of the map. This article proposes a new method to allow these people to be the designers of their own cognitive maps. A process, called synthesis, builds then a single cognitive map from this set of maps. The divergences in the maps due to the different points of view have to be solved. To do so, preferences on the designers are defined, they are used to favor the knowledge brought by some designer over the other ones.