| 摘要: |
| 本文针对基于Moore-Penrose广义逆实现的联想存储模型(如Kohonen模型、Mu rakami模型)缺乏对已存数据完全的联想回忆能力和非线性映射能力，通过在这些模型中引入一个扩展层(隐节点层)使原模型具有对已存数据的完全回忆能力和一定的非线性映射能力，通过矩阵的奇异值分解，从理论上阐明了改进模型的性能优越性．模拟结果证实了这一点． |
| 关键词: 广义逆，联想存储，神经网络，奇异值分解，非线性映射． |
| DOI： |
| 分类号: |
| 基金项目:本文研究得到国家基础研究“攀登计划”基金资助. |
|
| GENERALIZED INVERSE ASSOCIATIVE MEMORY WITH NONLINEAR MAPPING CHARACTERISTICS |
|
Chen Songcan,Gao Hang,Yang Guoqing
|
| Abstract: |
| Considering some AM(Associative Memories)(such as Kohonen and Murakami models)implemented with generalized inverse lack complete recall to the stored data and nonlinear mapping abilities，in this paper an improved associative memory model with complete recall and nonlinear mapping abilities is presented by adding one extended (hidden)layer to original AM models．The theoretical analysis based on matrix SVD(sin-gular value decomposition)and the simulation on computer show the performance superi-ority of the improved model． |
| Key words: Generalized inverse,associative memories,neural network,singular value decomposition,nonlinear mapping． |
