Maude Manouvrier

EDUCATION

Nov. 1996 - Jan. 2000

1996

1995

1993-1995

1991

RESEARCH INTERESTS

Transactional Web Service Composition and Content-Based Image Retrieval

RESEARCH ACTIVITY

Since 1996, I have been working on the management of similar large objects in databases, with Geneviève Jomier in the Databases and Software Engineering Group - LAMSADE and from 1998 until 2003 I have been working in an international CNRS -CONICIT cooperation with the team of Marta Rukoz, Professor at the Central University of Venezuela (accord numbers 5485, 7202, 8680 and 10058). Since 2009, I am working in international CNRS -CONICIT cooperation with the team of Yudith Cardinale, Professor at the Simon Bolivar University of Venezuela (accord number 22782).

During my doctoral studies, my research dealt with the management of similar large loosely or highly structured objects in databases. Then, my research interests was focused on indexing mechanisms for Content-Based Image Retrieval. Currently they are focus on Web Services for Transactional Composition.

A clear tendency observed in computer sciences is a shift from text data to multimedia, in particular images. This tendency concerns also databases. Data volume and availability more and more characterize today's databases. Crucial problems pointed out by all researchers concern the issue of organizing and manipulating this kind of data (e.g. storage, methods, data structure, retrieval and manipulation). This type of problems appears frequently in all kind of scientific applications -e.g. geography, medicine.

The main result of my thesis is the proposition of a set of structures that help managing similar large objects.

These kinds of objects appear in medical image databases. These images are processed in order to emphasize different image features facilitating medical diagnosis. As a result, one original image may give birth to many images successively processed according to different algorithms. Even more, an image may be decomposed into parts; the parts may be processed differently and eventually recomposed to emphasize some features. As a result, the database contains a great number of images that are interrelated and partially redundant. A database management system is required to support processing these images, while preserving their interrelations. In particular, it is required to efficiently support images comparison.

Images are large objects loosely structured. Quad-tree may encode them. We proposed a structure, called Generic Quad-Tree, is proposed in [JMR99, JMR00, MRJ02] (see publications) to store similar images. This work was done with Geneviève Jomier and in co-operation with the team of Professor Marta Rukoz from the Central University of Venezuela and was supported by the CNRS in France and the CONICIT in Venezuela. Two images are defined similar if, for both images, the quad-trees that encode them are similar, i.e. differ only for a relative small number of nodes. The Generic Quad-Tree optimizes the memory space of similar images and allows processing of operations on them, like images comparison, comparison of the same area in different images, or simultaneous updates in different images. An index is proposed in [JMO+04,JMO+05] for content-based image retrieval when images are organized by quadtree, and a survey on quadtree uses in image domain appears in [MRJ05] (see publications). Moreover, a generalized distance between images organized in quadtrees is proposed in [RMJ02,MRJ05b,RMJ06] (see publications).

PREVIOUS RESEARCH AREA

Content-based image retrieval integrating spatial relationships between image objects

Content-based image retrieval can be improved by integrating the spatial layout of objects in the image. Several approaches have been proposed for integrating spatial relationships into the image description. A study of the representation fo spatial relationships on symbolic images have been done in [GBM+08] (in french). Then, my research interst have been lied on the spatial information embedded into image content description for scene retrieval [HGB+10]. This work has been done with Nguyen Vu Hoang (LAMSADE), Valérie Gouet-Brunet (CEDRIC-CNAM) and Marta Rukoz (LAMSADE).

CURRENT RESEARCH AREA

Web Services for Transactional Composition

Transactional process in a peer-to-peer (P2P) requires algorithms to manage concurrent access to data and ensure a consistent process transactional. In such an environment, each pair is completely autonomous, can be both client and server and can appear or disappear at any time from the system, eliminating de facto the notions of master-slave or coordinator generally used in algorithms of distributed databases. Based on our experience in the field of nested transactions in distributed system, our goal is to offer models to control the concurrency and the QoS for Web services. The project is beginning. The first results appear in [EHM+07], EHG+08, EHM+10]. This work is done with Joyce El Haddad (LAMSADE) and Marta Rukoz (LAMSADE).

This work is currently supported by the Franco-Venezuelan CNRS-FONACIT Project with the research team of Yudith Cardinale from Universidad Simón Bolívar (USB). We currently address the issue of selecting and composing Web Services (WSs) considering functional and Quality of Service (QoS) requirements combined with transactional properties. In [CEH+10, CEH+11a], we formalize the WS functional, QoS, and transactional properties as well as the WS Composition problem, and we have extended Colored Petri Net (CPN) formalism to incorporate transactional WSs properties. This CPN-TWS selection algorithm that satisfies the user query functional conditions expressed as input and output attributes, QoS requirements represented by weights over criteria and, transactional properties expressed as a risk level. Another approach is presented in [BCV+10]. A state of the art on the composition of transactional Web services has also been presented in [CEH+11b] and in [CEH+13].

We have also oriented our work towards the application of operations research methods to the composition of transactional Web services, in collaboration with Virginie Gabrel and Cécile Murat, RO researchers from LAMSADE. This project was selected in 2010 as "Action de Recherche Fondamentale en Recherche Opérationelle" of the GDR RO (Group of french researchers in Operations Research). First results of this work was presented in a Master 2 training report on of Imen Megdiche about 0-1 integer programming for transactional web service composition (report written in french) in 2011 and published in [GMM+12]. Analysis of theorical complexity and new models were defined in [GMM13, GMM14, GMM15]. Our research project have been presented in API Days 2013 (Paris, Dec. 2013) -- to see the video.

With PhD students, we also proposes two frameworks: one for searching Data and Services with SPARQL [MGM14, MGM15] and one for for Transactional Service Selection Based on Crowdsourcing [AMR15].

SCIENTIFIC PUBLICATIONS

REVIEWER OR PROGRAM COMMITTEE ACTIVITIES