IRIT - UMR 5505

Français
CNRS
INPT
UPS
UT1
UTM
  Bandeau IRIT
 

  Sibertin-Blanc Christophe


  Status : Permanent
  Service / Team :  Systèmes MultiAgents Coopératifs
  Contact : sibertin@irit.fr
  Localization : UT1 - Manufacture des Tabacs, ME313
  Phone : 05 61 12 87 94
  Publications : Publications

  Professor in Computer Science at Université Toulouse 1 - Capitole

  Researcher at IRIT


 

 

Research

Recherche-en

 The MAELIA project: simulation model of a Socio-Ecological System

The MAELIA project ((Multi-agent for Environmental Norms Impact Assessment) develop a high-resolution  multi-agents simulation Platform for the study the environmental, economic and social impacts of various regulations regarding water use and water management in combination with climate change. This transdisciplinary project involves 4 Toulousan labs and it integrates a number of dimensions of socio-ecological systems such as hydrology, agronomy, farming, socio-economic processes (industrial and domestic water uses, land use, …), climat together with the management of water resources (regulatory constraints, management of dams, prefectoral decrees, …). Beyond the acquistion of a great variety of empirical knowledges, the elaboration of such a model raises questions regarding:
  • the need of a meta-model enabling the various scientific disciplines to integrate their specific knowledges into a homogeneous and shared representation of the system, and to implement this model in a simulator whose working is as transparent as possible;
  • the instantiation of this model from raw data that are incomplete and heterogeneous regarding space, time and functions;
  •  the calibration of the numerous processes which strongly interact.Publications

The SocLab project: Modelling of Social Organizations

Sociology of Organized Action (Crozier and Friedberg, 1977) is a sociology theory which intends to reveal the effective running of a social organisation, beyond its formal and explicit rules. This theory takes place in many courses, and underpins most methods for organizational diagnosis (analyse the cause of organisational dysfunctioning) or for change managing. It allows to understand how and why the members of an organisation use resources they master to maintain or increase their power and so keep some freedom of action, even limited, to achieve their own goals. This transdisciplinary project is carried out together with sociologist of Toulouse 1 University to formalise this theory.
The SocLab Platform is an organizational diagnosis tool allowing the user to:
  • model the structure of a real organization, according to the sociological analysis of the means and behaviors of its main actors, or of a virtual organization (thesis of Matthias Mailliard);
  • analytically analyse its structural properties (who could do what, conflicts, alliances, …) or properties of specific configurations (thesis of Paul Chapron);
  • compute, by a simulation algorithm that model the actors' bounded rationality for playing the so called social game, how they are likely to behave one another in terms of cooperation (thesis of Joseph El Gemayel).
Most recent works are about making the model of social actors behaviors to account for social emotions and their level of commitment (or withdrawal) into the organization.
From the computer science point of view, the importance of the organisational level in MultiAgents Systems is more and more recognised, and many works are inspired by Ethology. Sociology, and more precisely the Sociology of Organized Action, is a wealthy source of well-founded metaphors for models of coordination and organisations in MAS. In addition, techniques developed to model the behaviour of social actors could be profitably used to provide virtual reality characters, personal assistants or any kind of avatars with a human being-like social behaviour.
From the sociology point of view, the formalisation imposes a rigor that raises new questions and so leads to deepen this theory. For a discipline having no possibility to experiment, SocLab constitutes a virtual laboratory of a great scientific interest for the teaching, the diagnosis of unsafe organisations and the scientific analysis, comparison, assessment, etc. of organisational structures. Publication

 

Top

 

Meta-Modelling

The use of any modeling formalism induces a certain look on the system under consideration. Any model relies on a meta-model whose syntactic, semantic and pragmatic qualities are essential for both the model itself and for the modeling process. The underlying metamodel for Petri Nets with Objects (PNO) and Cooperative Objects (COO ) is relatively simple: the structure of a system consists of Entities, Actors and Operations, from which can be defined a set of Actions such as " an Actor performs an Operation with Entities". The behavior of a system, i.e. the set of all of the feasible trajectories in the state space, is then determined by its control structure that determines the allowable sequences of occurrences Actions  (the acts) resulting from the synchronization of the Actors' activities, system processes or Entities' life cycles. All this is very elegantly modeled with Petri nets that allow true concurrency and asynchronous interactions. Publication

 

Top

 

 Coordination in MAS

Most of the properties specific to MultiAgents Systems result from the “social capabilities” of agents, that enable them to intensively interact the ones with the others by means coordination, collaboration, cooperation, communication ... mechanisms. When agents interact in a way that is ruled by some constraints, these mechanisms are grounded upon the concept of protocol, as a type of flows of communication acts that are causally related.
The quality of the specification, design, validation and enactment of protocols is thus an essential part of MAS engineering; it determines the possibility to develop MAS that are, e. g.,  reliable, flexible, maintainable, scalable and supporting interoperability, even in an open environment. An essential challenge for MAS is to ensure both the autonomy of agents and the respect of the protocols’ rules.
We consider protocols as first class entities of MAS that must be treated as such at each level: specification, validation, software architecture and runtime. To this end, we have designed a meta-model of protocols that goes with a Petri net-based specification language, a component-based architecture allowing to use protocol as coordination resources, and an execution model where each conversation (i. e. instance, or occurrence of a protocol) is a process managed by a specific component. In addition, each agent has the possibility to synchronise its participations to simultaneous conversations without a full knowledge of the protocols’ rules.Publication

 Inter-organizational WorkFlow

The current trends of the production and service markets lead organisations (enterprises, administrations, associations, …) to build collaboration networks. To render a new service, that goes beyond the capacity of each organisation, they connect their Information Systems, gather their competences, share resources and so set cooperative information system that is distributed and heterogeneous. Inter-Organisational Workflow is the technology that aims at supporting this kind of activity and it addresses typical issues such as:
  • finding partners offering the needed abilities,
  • sharing information while keeping its own advances,
  • solving heterogeneity difficulties,
designing the composition of distributed processes and orchestrating their execution.
According to an Agent Oriented Information System approach, we have defined a coordination model including a component-based system architecture (ontology, matchmaker, …) together with a coordination specification language. Workflow services are specified by means of Petri nets with objects, so that they can be analysed and simulated, and then translated into OWL-S, so that they can be published and invoked as webservices. Deontic extension of Petri nets are under study.Publications

 Semantics for UML Sequence Diagrams

Sequence Diagrams (SD) are one of the UML diagrams used to model the dynamic dimension of a system. They allow to describe interactions between the system and the actors of its environment or between the component objects of the system. A SD describes the sequence of communications that occur in the course of a selected run of the system, that is the trace of the messages that are exchanged during this run. SD are a very popular language because of their clear graphical layout that gives an immediate intuitive understanding.
However, UML 1.x SD have no formal semantics since the dynamic of interaction is defined as a global ordering of the messages and not as an ordering of the sending and receiving actions performed by the objects. This results in an ambiguous reading of a UML 1.x SD: the representation suggested by the visual layout – the succession of events along the top-to-down lifelines synchronized by the horizontal messages –, that is not in accordance with the unique thread of control gathering all messages in a single sequence.
This lack of a formal semantics is one of the drawbacks of UML 1.x, and the same holds for its standardization as the ISO/IEC 19501 specification. It prevents to deduce from a SD specification the internal behavior of objects, the automatic generation of code, to have a global model of the system, and the use of SD as test cases for the functional validation; from a cooperative engineering point of view, it entails ambiguities in the understanding of SD and thus frustrates the UML objective to ease the communication between the stakeholders of a software project.
The 2.0 version of UML reduces this deficiency by providing SDs with a MSC-like operational semantics, without changing their graphical layout. This change raises questions about the reuse of SDs that have been earlier drawn in a UML 1.x perspective and thus require a new interpretation into the UML 2.0 semantics: to what extent the informal semantics that UML designers have given to UML 1.5 SD may be caught by the UML 2.0 formal semantics of SD?
We have addressed these issues by the proposal of four different operational semantics for UML1.x SD, based on various interpretation of the  following " For each message, its predecessors are the set of Messages whose completion enables the execution of the current message. All of them must be completed before execution begins ", so that designers can avoid the ambiguity of SDs. The respective relevance and the (structural and behavioural) properties of these semantics are studied. Once each SD of a specification has a well-defined semantics, it is possible to build the behavioural model of the whole system and notably to study its realisability.
The reference formalisms used in this work are partial order multisets (pomsets) and Petri nets.Publications

 

Top

 

Teaching

Software Engineering, in the M2 Ingénierie et Gestion des Systèmes d’Information
(Informatics and Society, Software Properties and Qualities, Software Processes, Specification, Design, V&V and Tests, Management of Delays and Costs)

Modelling of Social Systems, in the M2R Economie du dveloppement des Territoires

The R statistical software, in the M1 Toulouse School of Economics

IModlisation et Ontologies, in the M2 Droit et Informatique

The UML for Web Development, in the Licence Professionnelle Administration de sites Internet

 

Top

 

 Supervision

SocLab : un environnement pour la modélisation, l'analyse et la simulation des organisations sociales fondé sur la sociologie de l'action organisée : Matthias Mailliard (Thèse de doctorat, en cours)

Introduction du flou dans le modèle des organisations sociales selon Crozier et Friedberg : Vu Quang Anh NGUYEN (stage recherche M2 informatique, juin 2007)

Une plateforme pour jouer au jeu social : Fabien Lavigne, Antoine Letouzey (TER du M1 d'informatique, juin 2007)

Analyse des propriétés et des résultats de simulation dans SocLab : Damien JULIAN, Camille RICHARD, Fabien CROCHET, Sébastien LATRE, Hervé EICHWALD, Fabien BRICOUT (projet 3ème année ENSEEIHT, janvier 2007)

Etude des diagrammes de la dynamique d'UML par les réseaux de Petri : Omar TAHIR (Thèse de doctorat, décembre 2006)

 Sujets de stages de recherche

TITRE : Un modèle de la rationalité des acteurs sociaux orienté vers les équilibres de Nash

Nous avons entrepris depuis 4 ans, en collaboration avec des sociologues de l'Université Toulouse 1, une formalisation de la Sociologie de l'Action Organisée de M. Crozier et E. Friedberg. Cette théorie est largement enseignée et utilisée par les consultants appelés pour analyser l'origine des dysfonctionnements d'une organisation sociale. L'environnement SocLab (cf SourceForge) permet de décrire la structure d'une organisation, de simuler le comportement des acteurs sociaux (par apprentissage selon une rationalité limitée), et d'analyser les résultats produits dans le contexte de l'organisation étudiée.

Le modèle de rationalité qui est implanté dans SocLab est orienté vers la coopération, de façon que le jeu des interactions entre les acteurs se stabilise vers un optimum de Pareto (personne ne peut gagner plus sans que ce soit au détriment d'un autre). L'objet du stage est de trouver un modèle de rationalité des acteurs tel que le jeu se stabilise vers un équilibre de Nash (chaque acteur joue défensif, de façon s'assurer d'obtenir le maximum qu'il puisse quelque soit le comportement des autres). Eventuellement, des modèles d'autres rationalités pourront être recherchés, par exemple tel que le jeu se stabilise vers un état « égalitaire » (l'écart type entre les gains des acteurs est minimal), (anti-)éliste, etc.

TITRE : Introduction du flou dans la rationalité des acteurs sociaux

Nous avons entrepris depuis 4 ans, en collaboration avec des sociologues de l'Université Toulouse 1, une formalisation de la Sociologie de l'Action Organisée de M. Crozier et E. Friedberg. Cette théorie est largement enseignée et utilisée par les consultants appelés pour analyser l'origine des dysfonctionnements d'une organisation sociale. L'environnement SocLab (cf SourceForge) permet de décrire la structure d'une organisation, de simuler le comportement des acteurs sociaux (par apprentissage selon une rationalité limitée), et d'analyser les résultats produits dans le contexte de l'organisation étudiée.

En l'état actuel, la description de la structure d'une organisation nécessite, de la part de celui qui modélise cette organisation, de donner des valeurs numériques pour les taux d'enjeux et de solidarités des acteurs, et de définir des fonctions (linéaires, quadratiques, sigmoïdes, ?) pour décrire l'effet du comportement d'un acteur sur les autres. Pour le modélisateur, il y a une part d'arbitraire donner des valeurs précises, par exemple 0,2 de solidarité de A vers B plutôt que 0,15 ou 0,3. L'objet du stage est d'introduire le flou dans le modèle de la structure d'une organisation, et de présenter au modélisateur des résultats de l'analyse de cette organisation qui tiennent compte de cette « fuzzification ».

 

 Miscellaneous

Responsible for the Licence Professionnelle Technologies de l'information et de la communication en Petites et Moyennes Organisations

Participation to the IsyCri project

Participation to the Cosmagem project

Member of the Complex Systems Society

Member of the European Social Simulation Association

 

Top