Catálogo de publicaciones - libros

In this paper we present a framework called Simulación designed to drive multi-agent based participatory simulations on various scales. Such simulations are defined as simulations where human actors and autonomous agents play similar roles. With Simulación, each participant sits at a computer and interacts with an agent running on it. Humans and agents are paired and the agents can be considered as gates to the simulation, interacting together.

Using Simulación, we were able to conduct experiments over both local area networks and wide area networks. We were thus allowed to measure the requirements for participatory simulations led across laboratory and local area network boundaries. We also present how we started making the agents within these simulations as autonomous as possible, targeting large-scale and wide area experiments.

In MMAS-based biological system simulation, it is a challenging task to deal with numerous interactions among a vast number of autonomous agents. In our work, a hybrid massively multi-agent systems (MMAS) model is developed, and it incorporates the characteristics of cellular automaton (CA) and system-level mathematical equation modeling to simulate HIV-immune interaction dynamics. The mathematical equations are adopted within the site of a two-dimensional lattice. As the average high density, agent interactions can be calculated according to the equations without significantly affecting the performance of the systems studied. In the mean time, the CA model keeps the spatial characteristics of HIV evolution among the sites. The simulation based on the implemented MMAS discovers the dynamics of HIV evolution over different temporal and spatial scales, and reproduces the typical three-stage dynamics of HIV infection.

There are several image segmentation algorithms; each one has its advantages and its limits. In this work, we aim to use the advantages of two algorithms, in a massive multi-agents environment. We use the FCM (Fuzzy C-Mean) algorithm, to manage uncertainty and imprecision and the Region Growing algorithm, to act locally on the image. The massive multi-agents paradigm is then introduced into the region growing process in order to improve the segmentation quality. However in some cases some defaults appear in the segmented image, we propose then the use of a double predicate for the Region Growing algorithm, through a massive cooperative process, in order to improve the quality of the segmented image. Massiveness of the system allows for a better quality analysis.

This paper presents a mobile agent-based framework for ambient intelligence. The goal of the infrastructure is to provide people, places, and objects with computational functionalities to support and annotate them. Using location-tracking systems the infrastructure can navigate Java-based mobile agents to stationary or mobile computers near the entities and places to which the agents are attached, even when the locations change. The infrastructure enables application-specific functionalities to be implemented within mobile agents instead of the infrastructure itself. It maintains the locations of people and objects, including computing devices, and allows mobile users to directly access their personalized services from stationary computing devices in the environment or from their portable computing devices. This paper presents the rationale, design, implementation, and applications for our prototype infrastructure.

This paper presents a framework called Himalaya that allows to design mobile multi-agent systems (MMAS) deployed on several computers. An MMAS is a set of connected hierarchies of intelligent agents. Every agent ( a node) contains cognitive elements ( knowledge, goals, capabilities), processes and sub-agents. An agent is also mobile, he can move inside his hierarchy or to a remote one. In addition, an agent can dynamically acquire (kind of inheritance) intelligent and computational components from his sub-agents. The mobility and the inheritance as defined in our framework favor a dynamic adaptability and reconfiguring of systems in order to face the increasing complexity of distributed and cooperative applications.

We discuss how humans interact with the environment like mental and physical harmonization. Keyword is “resonance”. Each human has his own natural frequency, which is a metaphor for personality or daily habitual behaviors. In the proposed framework, each human behavior reacts the environment and the environment performs sensor-data mining and extracts each human’s natural frequency. The environment constructed from a multi-agent system is always watching humans, and when there is information to give one particular human, the environment interacts with him by using his natural frequency, so he can spontaneously and efficiently get the information from the environment. To achieve this, we set up several interaction devices between humans and the environment as well as various kinds of many sensors.

Goopas is a location-aware personalized information notification service coupled with automated ticket gates in railway stations. The service was provided to 40,000 users by an earlier system implemented on J2EE, but had performance problems, because location-aware systems must finish user tasks before the users move, and such personalized systems require handling a large number of users. Agent Server was used for Goopas to replace the original system. This paper describes how Agent Server was used for a real service that provides high performance and strong capabilities to 100,000 users.

Libraries are changing. The next stage library is supposed to be u-library (ubiquitous library) where each material is equipped with an RFID tag and an agent runs on it. Agents also run on counters, shelves, and other library equipments. They communicate each other and with book agents. The library agent systems are unique in the sense: (1) The number is large; thousands in very small libraries and millions in larger libraries. The total number will rise up to billion by considering the inter-library loaning system. (2) The supposed life span is very long, at least twenty years and hopefully more than one hundred years. Due to the rapid progress of information technology agents in a wide spectrum capability and functionalities form a massively multi-agent system even in one library. In this paper we propose a concept of delegate agent system in order to uniformly deal with such massively multi-agent systems.

Several attempts have been made to utilize a large scale multi-agent system for real-world information services in mobile/ubiquitous/pervasive computing environments. In our world, social knowledge, such as human relations and spatial semantics, plays an important role in communication and behavior. Such social knowledge should be incorporated into multi-agent systems to increase the effectiveness of real-world information services. This paper describes an approach for acquisition and utilization of social relationships among people. A web-mining approach is used to obtain a social network, especially that of researchers. We also address semantics of place.

We introduce two Japanese national projects on disaster mitigation which seriously involve massively multi-agent simulation system. We describe a simulation integration architecture that combines various natural phenomena and huge number of human activities to estimate and predict a complicated disaster progress and countermeasures effects. We describe the state of the art of our system design and implementation related mainly about multi-agent programming. Finally, we discuss about the rationale of massively multi-agent simulation for the purpose of disaster mitigation.