Entrepôts, Représentation et Ingénierie des Connaissances
Publications du laboratoire

Recherche approfondie

par Année
par Auteur
par Thème
par Type
--------------------
- Combining imperfect knowledge of maybe corrupted sources doi link

Auteur(s): Gavin G., Bonnevay S.

Conference: 2017 ISEA Asia Security and Privacy (ISEASP) (Surat, FR, 2017-01-29)


Ref HAL: hal-01807044,_v1
DOI: 10.1109/ISEASP.2017.7976997
Exporter : BibTex | endNote
Résumé:

Many data management applications, such as setting up Web portals, managing enterprise data, managing community data, and sharing scientific data, require integrating data from multiple sources. Each of these sources provides a set of values and different sources can often provide conflicting values. To discover the true values, data integration systems should resolve conflicts. In this paper, we present a formal probabilistic framework in the expert/authority setting. Each expert has a partial and maybe imperfect view of a binary target tuple b that an authority wishes recovering. The goal of this paper consists of proposing a multi-party aggregating function of experts' views to recover b with an error rate as small as possible. In addition, it is assumed that some of the experts are corrupted by an adversary A. This adversary controls and coordinates the behavior of the corrupted experts and can thus perturb the aggregating process. In this paper, we present a simple aggregating function and we provide a formal upper-bound over of the output tuple error expectation in the worst case, i.e. whatever the behavior of the adversary is.