Marion Créhange

Application of a symbolic-connectionist approach for the design of a highly interactive documentary database interrogation system with on-line learning capabilities

The NOMAD system is a documentary database interrogration system based on a symbolico-connectionist approach. NOMAD makes use of the synthesis capabilities and flexibility inherent in this type of approach to increase its processing power as compared as compared to existing systems while proposing new operating modes directly accessible to a large number of users: •NOMAD manages multiple synthetic type views on its documentary contents in the form of neural topographies acting as case-memories as well as elaborated thematic browsing tools. •NOMAD manages a session memory based on the neural model of the novelty detector with the following three functions: cumulative recording of user need, managing user contradictions and proposing new orientations. •NOMAD also has extended learning capabilities, enabling it to improve its performance in the long term. In the introduction of this article, we justify the modelling choices made for NOMAD. The system will then be described in detailed, stressing the new possibilities regarding help in query formulation and improvement in retrieval performance. The evaluation campaign and the numerous evolution perspectives provided by the model are described in the final section.

Semantics of user interface for image retrieval: possibility theory and learning techniques

A spark erosion machine having a machine table fixed to the machine upright and on which can be arranged a working container for the dielectric fluid by joining a vertically movable, cross-sectionally C-shaped component to an L-shaped component in longitudinal section. The L-shaped component comprises a fixed vertical wall element forming the rear wall of the working container and the machine table plate, which is detachably connected to the fixed wall element through a seal. The machine table plate is fixable in the reference plane of the machine table by means of rapid fastening and centering devices.

Machine Learning Techniques for Progressive Retrieval in an Image Database

Use of interactive image storage devices will become more and more widespread in offices, and this incites to develop different ranges of friendly applications concerning image databases. The common point of the applications we deel with is that they focus on progressive retrieval in an image database (for instance on videodisk) coupled with an alphanumeric database. The paper is based on two main ideas : deep interactivity and considering a retrieval process as a machine learning process.

Object Models and Methodology for Object-Oriented Database Design

Advantages related to object-oriented technology are at present well known. However, when designing, they must be considered as key objectives on equal terms with the description of the semantic and computer requirements of a problem. In this paper, we present a methodology for object-oriented database design that aims at reaching these objectives through stages of different abstraction levels (conceptual, logical, and implementation stages), the use of models and graphical notations. Moreover, since there is no standard object-oriented data model, we propose to construct during the conceptual and logical stages a database representation as complete as possible and independent on a specific OODBMS.