Catálogo de publicaciones - libros

In this paper we present the design of an agent-based simulation language called SOARS (Spot Oriented Agent Role Simulator). SOARS is designed to describe agent activities according to roles within social and organizational structures. Role taking processes can be described in our language. SOARS is also designed according to the theory of agent-based dynamic systems. Decomposition of multi-agent interaction is one of the most important characteristics of our framework. The notion of spot and stage gives spacial and temporal decomposition of interaction among agents. We apply our multi-agent framework to policy analysis of emerging virus protection in the case of SARS. In the latter part of the paper, we explain the implementation of the SOARS simulation platform. The simulation engine and related built-in functional objects are implemented in Java language. An application user can describe the agent-based simulation model only by writing script in SOARS script language without knowledge of Java classes. If needed, the user can customize the function of SOARS by implementing additional Java classes. The easiest way to customize is to develop special functional objects, and the SOARS platform will be the interface between such customized objects.

A financial market example illustrates how Interactive Evolutionary Computation (IEC) can be used for agent-based model inversion. The example shows that IEC can be used to discover a model that reproduces synthetic data generated with a very simple model.

The following research investigates communication among workers for AIDS-related non-governmental organizations (NGOs) in Hong Kong. The study seeks to learn to what extent full-time and part-time workers for those NGOs applying to the AIDS Trust Fund (ATF) discuss their prevention or treatment programs with their colleagues, and relates this exchange of information to the strategies that the NGOs use to fund and implement their programs. Ethnographic fieldwork contextualizes network models of communication and division of labor derived from quantitative questionnaires with a multidimensional scaling algorithm. Qualitative and quantitative evidence suggests that competition for funding and a cultural preference for hierarchy in Hong Kong discourage the sharing of ideas. Information exchange occurs primarily through strong ties within and between offices, and the Hong Kong Coalition of AIDS Service Organizations (HKCASO) emerges as the chief mechanism for sharing HIV/AIDS prevention strategies among the NGOs in Hong Kong.

In this research, we propose group dynamics that enhances the performance of multiagent systems in the Social Dilemmas. As the mechanisms for group formation, mutual choice mechanism for 2-IPD is extended to that for N-IPD. Furthermore, group split based on the metanorm of mutual choice is introduced. As a result of simulations with an evolutionary approach, we confirm the enhancement of the performance of the systems consisting of self-interest agents.

We model an online consumer-to-consumer (C2C) market by employing the prisoner’s dilemma. To discuss the characteristics of goods traded on a C2C market, we define temptation and contribution indexes based on the payoff matrix of the game. According to the results of the simulation with the model, we find that a positive reputation management system can promote cooperative behavior in online C2C markets. Moreover, we also find that such a system is especially effective for an online C2C market where expensive physical goods are traded, whereas a negative reputation management system is effective for an online C2C market where information goods are traded.

In this paper we explore a model of a team of intelligent agents constructing a shared interpretation of the state of their environment. Each agent is modeled as a constraint satisfaction network of the Hopfield (1982) type. We show that in a noisy environment communication intensity often has a non-monotonic effect on team interpretive accuracy. We also investigate how team communication can correct erroneous individual interpretations stored in agents’ memories — errors concerning what they know, not only what they perceive. We also compare the effects of different communication structures, and show that communication structure matters only when agents are cognitively heterogeneous (each has a different repertoire of interpretations in his memory), while it has only a minor impact on team performance when agents are homogeneous.

In response to the Kyoto Protocol, an international greenhouse gas (CO2) emission trading market is likely to be created. Computational and gaming simulations are expected to be able to reproduce complex phenomena in economics and society, and helps us to experiment with various controlling methods, to evaluate systematic designs, and to extract the fundamental elements which produce the interesting phenomena in depth analysis. In a series of works, we developed a simple agent-based simulation framework and its applications for economic and environmental studies including the international CO2 emission trading, and constructed a Web application for the gaming simulation of the emission trading with human players to investigate and improve the behavior model for agents. In this paper, we show results of our gaming simulation experiments performed in 2002 and 2003 as a part of a lecture course for environmental studies in university.

This article reports results of a series of U-Mart experiments as an educational courseware. The participants in the experiments were 22 third year university students and 20 computer programmed software machine agents. This series of the experiments is the first test which was performed systematically by more than 20 human trader agents. First, conditions of the experiments are explained; second, results of the series of the experiments are reported. Contents of the acquired results are as follows; settled profits, order volume, rate of contracted volume, volume of order, and position control. Though the obtained consequences may be definitive, they would enlighten tendencies and peculiarities of human conducts in financial market such as U-Mart system.

In problems with non-specific equilibrium, common in social sciences, the processes involved in learning mechanisms can produce quite different outcomes. However, it is quite difficult to define which of the learning mechanisms is the best. When considering the case of a cross-cultural environment, it is necessary to evaluate how adaptation to different cultures occurs while keeping, at some level, the cultural diversity among the groups. This paper focuses on identifying an evaluation criterion using a comparison of various learning mechanisms that can manage the trade-off between adaptation to a new culture and the preservation of cultural diversity. Results show that: (a) For small and gradual accuracy from a less accurate learning mechanism, there is a tiny reduction in the diversity while the convergence time drops rapidly. For an accuracy level close to the most accurate learning mechanism, a reduction of the convergence time can be minor, while the diversity drops rapidly; (b) The evaluation of learning mechanism that performs better for fast converging while simultaneously keeping a good diversity before the convergence was performed graphically.

This paper proposes an agent-based simulation model for analyzing adaptive processes of organizational structures to environmental changes. The model is based on a framework of computational organization theory (COT) and performed using the method of inverse simulation with genetic algorithm. The model specifies an activity process in organization that is composed of tasks, agents and environments. The roles of agents: normal and leader, are defined based on some basic tasks of agents: recognizing tasks from environment, sharing knowledge required for processing tasks, and coordinating tasks among agents. Organizational structures are distinguished according to the configuration of the roles of agents resolving the tasks recognized. Adaptation process is represented as learning process of agents’ internal models of environments. Environments are expressed as task generators and fall into 5 types. This paper shows some fundamental relationships between organizational structures and dynamically changing environments.