Jean-Pierre Georgé

The AMAS theory for complex problem solving based on self-organizing cooperative agents

In this paper, we present an approach for the design of complex adaptive systems, based on adaptive multi-agent systems and emergence. We expound the AMAS theory (Adaptive Multi-Agent Systems) and its technical working. This theory gives local agent design criteria so as to enable the emergence of an organization within the system and thus, of the global function of the system. We also present the theorem of functional adequacy witch ensures that a cooperative self organizing system performs a suitable work. Applications of this theory in the multi-agent system framework led us to define the architecture and a general algorithm for cooperative agents. The originality of our approach lies in the very generic manner our re-organization rules work and that they are completely independent from the function the system has to compute.

Criteria for the evaluation of self-* systems

In the last years, the growing complexity of the current applications has led to the design of self-adapting systems presenting self-* properties. These systems are composed of several autonomous interactive entities. They behave autonomously and present enhanced characteristics allowing them to handle dynamics coming from exogenous and endogenous changes. In this paper, we propose a set of criteria for the description and evaluation of the adaptive properties of such systems. They aim to provide a concrete mechanism to analyze the quality of the design of adaptive systems, to evaluate the effect of self-* properties on the performances and to compare the adaptive features of different systems. The criteria are grouped into different categories: methodological, architectural, intrinsic, and runtime evaluation. They have been identified and specified by analyzing several case studies, which address self-adaptivity issues through different approaches with different objectives in various application contexts.

Making Self-Organising Adaptive Multiagent Systems Work

The context of computational entities is rapidly changing: the development of artificial systems such as the Internet, ubiquitous computing, pervasive computing and autonomic computing mean that these entities have to cope with emergent phenomena arising in their environment. Rather than attempt to eliminate such emergence, we start to explore how this might be deliberately harnessed. That is, address how we might seek to engineer MAS with desirable emergent properties. To do this we discuss what emergence might mean in the context of MAS, and consider a class of such systems: Adaptive MAS (AMAS) that might be used to bring about such emergence. After reviewing the theoretical adequacy of AMAS systems we go on to sketch an approach to making them: namely by focusing the design effort on equipping each agent with responses to the non-cooperative situations it may encounter. This approach is illustrated in a simple but effective flood forecasting system called STAFF. Finally we discuss the expected benefits and difficulties inherent in this approach and the likely way forward.

Experiencing Self-adaptive MAS for Real-Time Decision Support Systems

Hydrological phenomena are often very dynamic and depend on numerous criteria. The STAFF software is an adaptive model for flood forecast based on self-organizing multiagent systems. It is operational since 2002 in the Midi-Pyrenees region in France. The aim of this paper is to show the relevance of our approach to model complex natural systems by focusing on the results, architecture and self-organization mechanisms of a real world application. The main idea is to let the artificial system self-adapt towards the adequate model by confronting it to real data, thus ensuring that the resulting model represents reality. Moreover, since the MAS is constantly adapting, we obtain a dynamic and autonomous system that can take into account any future dynamics (strong perturbations, sensor breakdowns...) and able to provide decision-makers with usable information anytime.

A Natural Formalism and a Multi-agent Algorithm for Integrative Multidisciplinary Design Optimization

MultiDisciplinary Optimization (MDO) problems represent one of the hardest and broadest domains of continuous optimization. By involving both the models and criteria of different disciplines, MDO problems are often too complex to be tackled by classical optimization methods. We propose an approach which takes into account this complexity using a new representation (NDMO - Natural Domain Modeling for Optimization) and a self-adaptive multi-agent algorithm. Our method agentifies the different elements of the problem (such as the variables, the models, the objectives). Each agent is in charge of a small part of the problem and cooperates with others to find equilibrium on conflicting values. Despite the fact that no agent of the system has a complete view of the entire problem, the mechanisms we provide allow the emergence of a coherent solution. Evaluations on several academic and industrial test cases are provided.

Towards an Adaptive Multi-agent System for Dynamic Big Data Analytics

The big data era brought us new data processing, data management challenges to face. Existing state-of-theart analytics tools come now close to handle ongoing challenges, provide satisfactory results with reasonable cost. But the speed at which new data is generated, the need to manage changes in data both for content, structure lead to new rising challenges. This is especially true in the context of complex systems with strong dynamics, as in for instance large scale ambient systems. One existing technology that has been shown as particularly relevant for modeling, simulating, solving problems in complex systems are Multi-Agent Systems. This article aims at exploring, describing how such a technology can be applied to big data in the form of an Adaptive Multi-Agent System providing dynamic analytics capabilities. This ongoing research has promising outcomes but will need to be discussed, validated. It is currently being applied in the neOCampus project, the ambient campus of the University of Toulouse III.

MAS4AT : un SMA auto-adaptatif pour le déclenchement d'alertes dans le cadre de la surveillance maritime

Cet article presente MAS4AT, un systeme multi-agent cooperatif et auto-adaptatif pour le declenchement d’alertes lors de la detection de comportements suspects dans le cadre de la surveillance maritime. Ce systeme est concu et developpe dans le cadre du projet europeen I2C qui vise a mettre en œuvre une nouvelle generation de systemes de surveillance maritime, capables d’aider les operateurs humains (i) a identifier les comportements anormaux de navires, (ii) a evaluer la suspicion associee a ces comportements et (iii) a declencher des alertes s’ils representent des menaces. Cet article introduit le projet I2C puis se consacre plus particulierement a la presentation de MAS4AT et a ses capacites d’apprentissage par renforcement.

AMAK - A Framework for Developing Robust and Open Adaptive Multi-agent Systems

Multi-agent systems are commonly used in various research fields such as artificial intelligence, operational research, simulation, biology, ... However, this diversity often requires that the system and agents in it are created from scratch for each new research project. In addition to the fact that this forces the developers to code similar elements anew each time, this can introduce non-negligible biases (e.g. an information accessible to every agent which shouldn’t be or a scheduler executing twice due to a user interface design failure). To avoid this, we propose in this paper AMAK, a framework developed in Java\(^\textsf {TM}\) to facilitate the design and development of a multi-agent system. First, we present the particularity of Adaptive Multi-Agent Systems. Secondly, a state of the art of the main tools and software aiming at facilitating the development of such systems is discussed. Then, we develop the architecture of the framework and the main features. The use of the framework is illustrated with an application for socio-technical ambient systems. And finally, we conclude with the perspectives of this work.

Autonomous Pareto Front Scanning using an Adaptive Multi-Agent System for Multidisciplinary Optimization

Multidisciplinary Design Optimization (MDO) problems can have a unique objective or be multi-objective. In this paper, we are interested in MDO problems having at least two conflicting objectives. This characteristic ensures the existence of a set of compromise solutions called Pareto front. We treat those MDO problems like Multi-Objective Optimization (MOO) problems. Actual MOO methods suffer from certain limitations, especially the necessity for their users to adjust various parameters. These adjustments can be challenging, requiering both disciplinary and optimization knowledge. We propose the use of the Adaptive Multi-Agent Systems technology in order to automatize the Pareto front obtention. ParetOMAS (Pareto Optimization Multi-Agent System) is designed to scan Pareto fronts efficiently, autonomously or interactively. Evaluations on several academic and industrial test cases are provided to validate our approach.