Multi-agent system description programming language system based on workstation cluster type parllel computer system

点击翻译按钮获取中文摘要

Multi-agent system is drawing attention as a structural model for many software systems including artificial intelligence systems. In multi-agent system, a number of autonomous agents cooperates mutually. Each agent is generated as a process or a task and can have its own situation and operates under the situation.Situation consists of states and procedures where both of them can be dynamically changed in general. Therefore, it becomes necessary for the agent to dynamically hold the states (data) and procedures (hereafter we refer them as a knowledge base).Moreover, when a number of agents are cooperating, it is necessary for them to share knowledge bases and conduct knowledge communications between agents. We proposed a computation model for cooperative agents. In this model, we introduce a field as an abstraction of the group of agents. Each agent can communicate to the other agents in the same group through a field. The field also enables us to specify the action of the group, i.e., … More it is used for a knowledge base to keep rules and data shared by agents in the group. Based on this model, programming language which is based on Prolog was designed and its prototype processing system was built on the network connected distributed multi-workstation system. We also developed a parallel processing system for analyzing phylogenetic relationships of mircroorganisms based on a maximum likelihood method. Methods for inferring relationships from molecular sequence data are especially valuable, given the enormous increases in DNA sequence data. The maximum likelihood method uses concrete models of the evolutionary process and are well-motivated statistically, but the computational cost has hindered the use of this method for inferring trees with more than about 20 organisms. We parallelized the maximum likelihood method by utilizing two types of parallelism, parallel evaluation of phylogenetic trees and parallel computation of likelihood values. By combining these two parallelisms, we obtained significant speedup. Less