分布式多Agent局部控制下形成的宏观自组织

集中式协作过滤算法中 ,服务器的负荷过大且成为瓶颈环节 ,该文研究了分布式算法。Agent只有局部视角 ,算法以有限朋友列表和信任度为基础。文中将信任度的控制规则分为比例、积分、微分规则。通过仿真实验研究了比例、积分规则对系统全局性能的影响。实验结果表明 ,各个Agent通过自适应学习 ,逐渐与自己的朋友形成了恰当的信任联系 ,该连接强度反映了合作的密切程度。同时 ,所有这些连接强度构成的加权连接图 ,反映了多 Agent的宏观聚类效果和自组织现象

Macro self-organization based on distributed multi-agent control in local view

With centralized collaborative based information filtering, the central node becomes the bottleneck. A distributed multi agent collaborative based information filtering method is proposed based on the limited friends list and trust relationships. The agent has only a local viewpoint and changes the trust value through interaction. The trust update control rules are classified into three classes, which are proportion, integral, and differential coefficient (PID). Our simulation experiments used proportion and integral control rules. Through learning, the agent learns the righter and righter trust value to its friends. The ranked connectivity of different agents forms a ranked graph which also reflects macro clustering and macro self organization processes.