Sabine Moisan

Recent papers

Design of Knowledge-Based Systems

• Une plate-forme pour une programmation par composants de systèmes à basede connaissances/A Platform for Component Programming of Knowledge-Based Systems

Sabine MOISAN,
Habilitation à diriger des recherches en Sciences, Université de Nice-Sophia Antipolis, avril 98
Résumé/Abstract:
Ce manuscrit décrit les travaux de recherche que j'ai effectués dans le domaine de l'intelligence artificielle, et plus particulièrement des systèmes à base de connaissances (SBC). L'utilisation des systèmes à base de connaissances est freinée par deux sources de difficultés importantes : difficulté informatique dans la construction de moteurs de SBC bien ciblés, et difficultés dans la réalisation de bases de connaissances pour des experts des dommaines d'application. Pour répondre à ces problèmes, quatre axes de recherches ont été explorés : réalisation de systèmes plus génériques, extensions des capacités d'expression des connaissances, amélioration de l' intégration des SBC et de leur communication avec d'autres logiciels, et enfin rapprochement avec les méthodes et techniques du génie logiciel. Le manuscrit expose les travaux réalisés dans le cadre des évolutions de deux générateurs de SBC : Smeci, dédié à la tâche de conception et Ocapi (et sa variante temps-réel Planete), dédié à la tâche de pilotage de programmes. Enfin, pour mieux répondre à la nécessité d'évolution des générateurs, une plate-forme logicielle plus générique a été développée. Cette plate-forme, nommée Lama, fournit un ensemble d'outils unifiés aussi bien pour pour construire des moteurs de générateurs, que pour développer des SBC. Les principaux outils fournis sont : un langage de représentation des connaissances, un vérificateur de bases de connaissances et une bibliothèque de composants pour la construction de moteurs. Cette plate-forme est un premier pas pour résoudre les difficultés des concepteurs de moteurs et des experts et leur apporter plus de souplesse. Elle a été utilisée pour la réalisation de différents générateurs de SBC en pilotage de programmes, qui ont eux-mêmes été appliqués au pilotage de traitement d'images, par exemple en télédétection ou en imagerie médicale.

This manuscript describes my research in the area of artificial intelligence, and particularly on knowledge-based systems (KBS). Two important reasons inhibit the widespread utilization of knowledge-based systems: first, computer science programmers have difficulty designing tailored KBS engines and second, experts in a real-world application domain have difficulty building a knowledge bases. We have pursued four different approaches to remedy this situation: designing more generic systems, extending the capacity of knowledge expressiveness, improving the integration of KBSs and their communication with other software, and finally exploring the connections with software engineering methods and techniques. The manuscript describes improvements incorporated during the evolution of two generators of knowledge-based systems: Smeci, which was dedicated to the design task and Ocapi (and its real-time variant Planete), which was dedicated to the task of program supervision. Finally, to better address the necessary evolution of KBS generators we developed a more generic software platform. This platform, named Lama, provides designers and experts with a set of unified tools for designing generator engines, as well as for developing entire KBSs. This tool set consists mainly of a knowledge representation language, a knowledge base verification tool, and a library of components for engine design. The platform is a first step to overcome the difficulties of both engine designers and experts, and to provide them with flexible tools. It has already been used to design several KBS generators in program supervision, which in turn have been applied to supervise image processing programs for e.g. teledetection or medical imaging.
 

• Experiments in Building Program Supervision Engines from Reusable Components

Monica CRUBÉZY, Mar MARCOS, Sabine MOISAN
13th European Conference on Artificial Intelligence (ECAI'98). Workshop on Applications of Ontologies and Problem-Solving Methods. August 1998, Brighton, UK.
Abstract:
In this paper, we present a case study of reasoning engines built thanks to our knowledge-based system (KBS) development platform LAMA. The LAMA platform supports the software-level composition of KBS engines for a given task from reusable reasoning blocks. This paper emphasizes engine design activity and shows how our platform helps easily create new engines or modify existing ones. Indeed, designers need help to build the problem-solving method (PSM) of an engine, appropriate for an application domain. The current available tools neither incorporate designers' knowledge at the right level, nor offer development, integration and testing facilities to implement KBS engines. Our experience tackles problems such as PSM and ontology design and (partial) reuse. This involves aspects such as the choice of an appropriate set of reasoning blocks to cover the needs of a problem-solving task, with the right level of granularity. As a first experimental field, we focus on program supervision (PS), an example task in which we have several previous experiences. PS aims at automating the management of expert decisions about the use of programs, that are difficult to handle by non-specialists. We report here our experience in building three PS engines by means of the LAMA platform. This experiment brings new insights in the domain of component-based approaches to PSM design.
To get this article,click here

Program Supervision

• A Model-Based Approach to the Verification of Program Supervision Systems

Mar MARCOS, Sabine MOISAN, Angel P. del POBIL.
4th European Symposium on the Validation and Verification of Knowledge Based Systems (EUROVAV'97) - Leuven, Belgium. June 1997. Proceeding pp. 231-241.
Abstract:
This paper is concerned with the verification of knowledge-based systems for a task called program supervision (PS). Our aim is to highlight how the model on which a particular PS system is based influences the properties that the knowledge base must fulfil. Adopting this point of view allows us to go beyond verification techniques that do not exploit the intended use of the embodied knowledge. We distinguish the following elements in the specification of a knowledge-based system: task definition, problem-solving method, and domain model. Our verification approach is based on the PS ontology and on the requirements that both the PS task and problem-solving method enforce on the domain knowledge of a specific application. The definition of the verification issues for the knowledge base is done in a way dependent both on the PS task and the particular PS problem solving method, but independently of the knowledge representation scheme or the specific application. This work in the modelling and property determination for the PS task has as additional outcomes a categorisation of the properties and some guidelines to determine the adequacy of the expert's knowledge with respect to the requirements of the PS problem-solving method. We also intend to rely on this approach in our future work on the verification of the PS problem-solving methods.
To get this article,click here
 
• Managing Complex Processing of Medical Image Sequences by Program Supervision Techniques

Monica CRUBEZY, Florent AUBRY, Sabine MOISAN, Virginie CHAMEROY, Monique THONNAT, Robert DI PAOLA
SPIE international symposium Medical Imaging 1997 (MI'97). Proceeding 3035-85 pp. 614-625. Newport Beach, California, USA. February 1997.
Abstract:
Our objective is to offer clinicians wider access to evolving medical image processing (MIP) techniques, crucial to improve assessment and quantification of physiological processes, but difficult to handle for non-specialists in MIP. Based on artificial intelligence techniques, our approach consists in the development of a knowledge-based program supervision system, automating the management of MIP libraries. It comprises a library of programs, a knowledge base capturing the expertise about programs and data and a supervision engine. It selects, organises and executes the appropriate MIP programs given a goal to achieve and a data set, with dynamic feedback based on the results obtained. It also advises users in the development of new procedures chaining MIP programs. We have experimented the approach for an application of Factor Analysis of Medical Image Sequences as a means of predicting the response of osteosarcoma to chemotherapy, with both MRI and NM dynamic image sequences. As a result our program supervision system frees clinical end-users from performing tasks outside their competence, permitting them to concentrate on clinical issues. Therefore our approach enables a better exploitation of possibilities offered by MIP and higher quality results, both in terms of robustness and reliability.
To get this article,click here
 
• Knowledge Modeling of Program Supervision Task

Mar MARCOS, Sabine MOISAN ,Angel P. del POBIL
11th International Conference on Industrial and Engineering Applications of Artificial Intelligence and Expert Systems (IEA-98-AIE). Benicassim, Spain. June 1998. Proceeding Springer Verlag - Lecture Notes in Computer Science 1415. Vol I, pp. 124-133 76.
Abstract:
This paper presents a knowledge-level analysis of the task program supervision based on two different systems: PEGASE and PULSAR. A knowledge-level analysis of a knowledge-based system reveals the organisation of the knowledge it uses, and how it uses this knowledge to solve the task. It is also the key to determine the set of properties that it assumes on domain knowledge. These aspects have been successfully used as a framework to compare different systems, mostly for knowledge engineering purposes. Our purpose is to use assumptions as the properties that the knowledge base must verify.
To get this article,click here
 
• What can Program Supervision Do for Software Re-use?

Monique THONNAT, Sabine MOISAN
International Conference on Industrial and Engineering Applications of Artificial Intelligence and Expert Systems (IEA-98-AIE).
Benicassim, Spain. June 1998. Proceeding Springer Verlag - Lecture Notes in Computer Science. 1416 Vol. II, pp 403-414 Use of Knowledge-Based Control for Vision Systems Experience in Integrating Image Processing Programs
Abstract:
We propose a framework for the development of vision systems that incorporate, along with the executable computer algorithms, the problem-solving knowledge required to obtain optimal performance from them. In this approach, the user provides the input data, specifies the vision task to be performed, and then provides feedback in the form of qualitative evaluations of the results obtained. These assessments are interpreted in a knowledge-based framework to automatically select algorithms and set parameters until results of the desired quality are obtained. This approach is illustrated on two real applications, and example from the knowledge bases developed are discussed in detail.In this paper we are interested in knowledge-based techniques (called program supervision) for managing the re-use of a modular set of programs. The focus of this paper is to analyse which re-use problems program supervision techniques can solve. First a general definition for program supervision and a knowledge representation model are proposed. Then an analysis is presented in terms of the structure of the programs to re-use and in terms of the effort for building a program supervision knowledge base. This paper concludes with what program supervision can do for program re-use from the points of view of the code developers, the experts, and the end-users.
To get this article,click here

Abstract:
We propose a framework for the development of vision systems that incorporate, along with the executable computer algorithms, the problem-solving knowledge required to obtain optimal performance from them. In this approach, the user provides the input data, specifies the vision task to be performed, and then provides feedback in the form of qualitative evaluations of the results obtained. These assessments are interpreted in a knowledge-based framework to automatically select algorithms and set parameters until results of the desired quality are obtained. This approach is illustrated on two real applications, and example from the knowledge bases developed are discussed in detail.
To get this article,click here

Abstract:
In this paper we are interested in knowledge-based techniques (called program supervision) for managing the re-use of a modular set of programs. The focus of this paper is to analyse which re-use problems program supervision techniques can solve. First a general definition for program supervision and a knowledge representation model are proposed. Then an analysis is presented in terms of the structure of the programs to re-use and in terms of the effort for building a program supervision knowledge base. This paper concludes with what program supervision can do for program re-use from the points of view of the code developers, the experts, and the end-users.
To get this article,click here

Abstract:
Fields such as image processing, signal processing or scientific computing have produced a large number of powerful programs and many different libraries of data processing programs have been developed. But few support is provided to new-comers or inexperienced users. Thus, for these programs to be widely applied, there is a need for an intelligent support. To help users of program libraries, we propose a knowledge-based approach, that we call program supervision. A program supervision system takes in charge the management of the library use i.e. the (re)configuration of data processing programs. Depending on the context of execution, a program supervision system has to take into account various constraints on resources, e.g. hardware, software or time. A typical case is real-timeapplications, where data flows coming from sensors have to be processed in limited time and under hardware constraints. This paper summarizes our program supervision approach and focuses on how program supervision can take into account resource constraints, and especially time constraints. A real-time experiment in program supervision illustrates this problem.
To get this article,click here

Abstract:
In this paper we present a study of verification and validation problem in the framework of knowledge-based program supervision systems, i.e. intelligent environments dedicated to the task of automation of complex software utilization. We are interested in methodologies and tools which can be applied during the knowledge-based system development process, in order to guarantee a certain degree of reliability in the final system.
To get this article, click here

Abstract:
Image processing in the fields of medical diagnosis and therapy follow-up is a challenging issue because of the complexity of the knowledge involved in describing the data and reasoning mechanisms a radiologist uses. A methodological radiologist would be helped if a system could schedule and execute operations and reason on data and user-goals, leaving him with the result interpretation. In this article we show what knowledge is needed to perform the supervision (that is planning and control of execution) of medical image processing programs, and how this knowledge should be described and structured. Our aim is to develop a program supervision tool to help a user in the manipulation of data and programs specific to medical imagery. It must therefore offer a framework to describe knowledge on data, programs and reasoning processes, as well as planning, execution and control mechanisms adapted to the expert in medical image processing. For this purpose, we have developed a constructive modelisation for data description that enables encapsulating knowledge where it is useful for program supervision and we propose a hybrid planning strategy.
To get this article, click here

Abstract:
Program supervision consists in different phases, that can be completely or only partly automated: planning, execution and control of result, evaluationand repair. This last phase has seldom been fully developped in a system. In this paper, we first analyze the most common needs for a complex repair mechanism, by surveying the facilities provided by existing systems. The failure handling mechanism involves three sub-mechanisms: plan repair, adaptation and clue propagation. For these sub-mechanisms, we propose a set of general functions to design hiher level repair functionalities for the experts.
To get this article, click here

Abstract:
This paper describes the use of a knowledge-based system to manage a library of image processing programs which performs the computation of land cover mixing in pixels of a multi-spectral satellite image. This system has been developed to help naturalists, such as geologists, pedologists, or foresters who are not specialists in computer vision. It automatically processes the data coming from satellites. The role of such a system is to convert the requirements of a user, expressed in terms of user requests, into the correct image processing commands. Consequently, the users can concentrate their attention on the interpretation of the results and not on the data management. Finally, some examples are shown.
To get this article, click here

Meta Programming and Reflection

Abstract:
In this paper we focus on the benefits that a metaobject protocol may bring to a C++ class library. This library, named FRAMELIB, implements "frames" as used in Artificial Intelligence. It addresses four important aspects: frame structure and behavior, type safety, frame instance persistence, and sophisticated debugging mechanisms. Although frames may appear rather similar to classes as found in object-oriented languages, their implementation in C++ raises a number of technical challenges. The interface to our library introduces protocols and patterns which must be strictly followed to ensure the correctness and consistency of the code. To simplify the usage of the library, we decided to recourse to a metaobject protocol (MOP), and more specifically to OpenC++. This paper describes our experience with this MOP, the promising results that we obtained, the problems we encountered, and the design issues that this technique raises.
To get this article, click here