MA4CS07

Over the last decade, Swarm Intelligence approaches to the engineering of complex and distributed software applications as well as multi-robotic systems have become more and more sophisticated and accepted. These approaches seek to design emergent functionalities of a complex system through the self-organizing dynamics of relatively simple interactive agents that are embedded in a shared physical or virtual environment. Traditionally, the focus of these swarms of simple agents has been the operation of a single complex application, but in several domains we have found that we need to model or control the behavior of many complex entities at a level of detail where a representation of each entity as a simple agent in a swarm is not powerful enough.

This talk presents the polyagent modeling construct that combines the power and simplicity of swarming agent systems with the explicit representation of complex interacting entities. In these models, each entity in the corresponding real world application is represented by a polyagent. A polyagent is in itself a multi-agent system comprising a single (complex) avatar agent controlling its own swarm of behavioral ghost agents. As ghosts of different polyagents interact with each other through the shared environment of a dynamic simulation, complex entity reasoning and interactions emerge. I will discuss the general design and implementation of polyagent systems as well as their use in specific applications.

Giuseppe Vizzari
(09:30-10:20)

Title : "Situated Multi-Agent Models for Complex Systems: the Cases of Crowds, Web Sites and Context Aware Ubiquitous Systems"

Abstract :

Several informal definitions of the term “complex system” emphasize the presence of interconnected (and often simple) parts whose interactions generate a higher order behaviour. The multi-agent systems paradigm seems thus a suitable conceptual, modeling and computational instrument to represent and reproduce “in silico” specific dynamic views of a complex system. On the other hand, MAS can also be a suitable approach to the design and implementation of artificial systems characterized by some specific features of complex systems. In this framework, this talk will discuss three case studies in which situated multi-agent systems were applied to represent and simulate complex systems (crowds of pedestrians) and to realize computer systems that, by interacting with human users, exhibit complex adaptive overall behaviours (adaptive web sites and context aware pervasive computing systems).

David Chavalarias
(11:00-11:50)

Title : "Endogenous distributions and strong emergence in social simulations"

Abstract :

Multi-agents modelers recurrently face the problem of the choice of their parameters’ values while most of them are exogenous. In presentation we will address the issue of endogenization of these parameters when it makes sense in a social learning perspective within the formalism of metamimetic games. We first show how its is possible to endogeneize the distribution of agents’ rule of behavior. Then we apply the method to endogenization of time constants in the model, each agent having its own subjective perception of time. In this perspective, the values of endogenous parameters are the outcome of a dynamical process characterized by agent’s cognitive capacities and environmental constraints. Last we will link this approach to the issue of strong emergence in multi-agent systems, described as emergence of a new perceptive category for agents, integration of this category in new rules of behaviors and finally modification of the overall distribution of rules of behavior in the population.