多重加权改进型指数双向联想记忆网络及其决策性能  

Multiple Weighted Improved Exponential Bidirectional Associative Memory Model

作　　者：陈松灿[1,2] 蔡骏[1] 

机构地区：[1]南京航空航天大学计算机科学与工程系 [2]南京大学计算机软件新技术国家重点实验室,江苏南京210093

出　　处：《电子学报》2002年第8期1200-1203,共4页Acta Electronica Sinica

基　　金：国家自然科学基金 (No .6970 1 0 0 4 ) ;南京大学软件新技术国家重点实验室基金项目

摘　　要：CCWang等作者利用指数双向联想记忆模型 (eBAM) ,构造了由多个eBAM构成的多重eBAM(Multi e BAM)信念组合模型 ,使之可模拟多个专家的投票表决决策 ,并获得了Multi eBAM在各eBAM具有同等权威度条件下的决策性能 .本文在此基础上 ,通过对各eBAM引入不同的权值来模拟各专家不同的权威度 ,推广了Multi eBAM .进一步借助陈所提出的改进型eBAM(IeBAM) ,构建了相应的多重加权改进型eBAM(Multi WIeBAM)信念组合模型 ,获得了此推理模型在同、异步方式下的决策性能及多专家不同权威度下的多数投票因子 ,使之更符合实际的多数表决决策 .理论分析表明Multi WIeBAM所获得的多数投票因子优于Multi WeBAM的多数投票因子 ,即前者较后者具有更紧致的下界 .实验结果也表明了Multi WIeBAM的性能要优于Multi WeBAM .C C Wang and coworkers built a belief combination model consisted of multiple exponential bidirectional associative memory(Multi-eBAM) through using eBAM with equal privileges and applied it into the voting decision making of multiple experts and obtained the latter decision-making performance. Using Chen improved eBAM (IeBAM) and endowing different privilege to each IeBAM or expert, a multiple weighted belief combination model consisted of IeBAM (Multi-WIeBAM) is constructed and investigated and becomes an extension to Wang Multi-eBAM model. Then its stability in synchronous and asynchronous updating modes in respectively proven and its corresponding decision-making performance and majority factor for different privilege of each expert are obtained. So the proposed model confirms the real-life voting decision. The initial analysis indicates that the majority factor of Multi-WIeBAM is tighter than that of the corresponding Multi-WeBAM, in other words, the former has better decision-making performance than the latter. Finally the experimental results also verify the above points.

关 键 词：决策 多证据推理 加权 联想记 神经网络 eBAM 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]