A112 エージェント社会における自己組織化現象(集団行動のマルチエージェントシミュレーション)

概要

論文の詳細を見る
There are many researches on emergent properties and collective phenomena. The previous researches have mainly focused on the global interaction or random interactions of agents. In this paper, we focus on local interactions of agents. We address the following main question : how do heterogeneous agents generate emergent properties? These questions will depend crucially on how they interact and adapt their behavior. Agents myopically evolve their behaviors based on their own rules, which is given as the function of their idiosyncratic utilities and the action of their neighbors. We show that agents' behaviors reflecting their micromotives combined with the behavior of others produce orders and sometimes unexpected phenomena. The heterogeneity of agents is main concern for emergent behavior. We consider the relation between emerged order and the diversity of the population of agents. Another heterogeneity of agents is concern of this paper. In this paper, we consider the collection of agents with optimizers or imitators. Optimizers are myopic best-response agents to the previous opponents' output. Imitators mimic the output decision of the most successful firms of the previous round. We show the conditions that a population of heterogeneous agents converges to more efficient state. The emerged collective phenomena cannot always be utilitarian for the each agent and efficient for the population. So, in this paper, we focus on the agent's utility and the efficient collective behavior. In local interaction of agents, location of heterogeneous agents in the population is important and we also characterized the configuration of agents, which is localization, as the diversity of the population. Here, we classify localization of agents into structural assignment and random assignment. Structural assignment is that agents with similar payoff parameter are located at nearby site. From the viewpoint of the population, it is well ordered. One the other hand, random assignment is that agents with similar payoff are located at random. And population is disordered. Our interest lies with systems in which many connected processors, with no central control, locally interact to procedure globally coordinated behavior. From the simulation results, we found that the structural assignment is effective for the both each agent and the population. And we found population of Imitator suffers from the localization of agents. So, we recommend the population of agent with Optimizer for the coordinate conditions. Besides, we found the research of the collective properties with global interaction of agents is helpful for the anticipating the collective behavior of local interaction.
一般社団法人日本機械学会の論文
2001-11-14

A112 エージェント社会における自己組織化現象(集団行動のマルチエージェントシミュレーション)

スポンサーリンク

概要

著者

関連論文

スポンサーリンク