par- 12 septembre 2011
Bonjour tout le monde
Je vous écris afin de vous informer que le mercredi le mercredi 7 septembre à 10h30 j’ai invité M. Daniel dos Santos Kaster de l’Université de Sao Paolo pour nous présenter un de ses travaux de recherche qui s’intitule "Similarity Queries in Relational DBMSs" (voir le résumé plus bas).
Abstract : With the increasing dissemination of devices to capture digital data, such as digital cameras and medical scanners, huge amounts of complex data are being generated every day. The main approaches to query complex data are to employ metadata-based retrieval, when metadata are available, and/or content-based retrieval. Content-based retrieval aims at automatically extracting features that describe the complex data content and use them in place of the data to answer similarity queries. Similarity queries have been widely studied in the past few years, however, the existing Database Management Systems (DBMSs) do not provide support for them yet.
In this presentation, it will be shown a solution attached to an off-the-shelf DBMS that adds to it capabilities to answer similarity queries. This solution permit to mix conventional metadata-based operators as well as similarity-based operators in the same query sentence, thus allowing enumerating alternative query plans for query optimization. It will also be presented a new variation of the k-Nearest Neigbor (k-NN) similarity query, which allows providing a condition constraining the search. This variation, called the constrained k-NN query, allows answering types of queries that cannot be represented using the conventional similarity operators.