论文部分内容阅读
本文讨论了不分层的通用前馈网络（GFFN），并提出了一种作为模式分类器的排序学习前向掩蔽（SLAM）模型及其算法实验结果表明，这种网络作为模式分类器用时，学习时间远小于各种改进的BP网络而且所需使用的神经元数量也有显著的减少本文还介绍了这种SLAM模型在应用双阈值神经元DTN时进一步减少神经元数量的实验结果及其网络结构和学习算法，以及这种模型的模式分类器所具有的不断扩展与改善的能力论文还介绍了SLAM模型模式分类器在CASSANDRA-I小型神经计算机上实现的实验结果：在256维输入空间1024个随机样本的分类情况，学习时间约3小时20分，判别时间为0．007秒．
This paper discusses the non-hierarchical universal feedforward network (GFFN) and proposes a sort learning forward mask (SLAM) model as a pattern classifier and its algorithm. Experimental results show that this network as a model classifier, The learning time is far less than that of various improved BP networks and the number of neurons needed to be significantly reduced. This paper also presents the experimental results and network structure of this SLAM model to further reduce the number of neurons when applying dual threshold neuron DTN And learning algorithms, as well as the ability of the model’s classifier to continuously expand and improve. The paper also presents the experimental results of the SLAM model classifier implemented on the CASSANDRA-I small neuron computer. In the 256-dimensional input space, 1024 Random sample classification, learning time of about 3 hours and 20 minutes, the discriminant time of 0.007 seconds.