受监督SOM学习模型定量输入-输出摄动分析

• 作 者: 戚湧 於东军
• 机 构: 南京理工大学计算机科学与技术学院,南京,210094
• 出 处: 系统仿真学报
• 发表日期: 2010年5期
• 关键词: supervised SOM perturbation analysis learning model neural network

给定一个学习模型,输入摄动会引起相关的输出摄动.然而,在受监督自组织学习模型中,此结论并不必定成立.针对受监督自组织学习模型的这一特性,提出一种基于统计的方法用于定量输入-输出摄动分析.所提方法对于该学习模型的精度分析及结构设计具有指导意义.数字仿真结果验证了所述方法的有效性及正确性.“,”Input-output perturbation analysis is one of the most important aspects in designing and analyzing a learning model. In many learning models, an input perturbation will often give rise to an associated output perturbation. However, this is not necessarily the case for a supervised self-organizing map (SOM) learning model. A statistical-based method was proposed to quantitatively analyze the inputoutput perturbation of the supervised SOM learning model by considering this special characteristic. The method will make contribution to both the accuracy analysis and the structure configuration of the supervised SOM learning model. Simulation results demonstrate the correctness and effectiveness of the proposed method.