论文部分内容阅读
在构建神经网络时,采集的训练模式总存在摄动,如何度量这种摄动,提出模糊集摄动度量的新方法。众多学者研究的两类形态学联想记忆网络的存储能力和抗腐蚀/膨胀噪声的能力等性质几乎都相同,但基于这种模糊集摄动的度量方法,研究训练模式摄动对两类模糊形态学联想记忆网络的影响时发现,两类网络对训练模式摄动的鲁棒性差异很大,其中一类模糊形态学联想记忆网络对训练模式摄动拥有好的鲁棒性;而另一类模糊形态学联想记忆网络的这个性质较差。研究内容对形态学联想记忆网络的性能分析、学习算法的选择和训练模式获取设备精度的选择有一定的指导意义。
In constructing neural networks, the training patterns collected always have perturbations, how to measure this perturbation, and propose a new method of fuzzy set perturbation measure. The properties of memory capacity and anti-corrosion / expansion noise of the two kinds of morphological associative memory networks studied by many scholars are almost the same. However, based on the measurement method of this kind of fuzzy set perturbation, When learning the influence of associative memory network, we found that the robustness of the two types of networks to the perturbation of training patterns is quite different. One type of fuzzy morphological associative memory networks has good robustness to training pattern perturbation; This property of the fuzzy morphological associative memory network is poor. The content of this study is of guiding significance to the performance analysis of morphological associative memory networks, the selection of learning algorithms and the selection of training accuracy for equipment acquisition.