摘要
分析了自组织神经网络各种改进算法的优缺点,详细设计和实现了一种基于改进动态二叉树的自组织映射树(DBTSONN)。在改进动态二叉树中神经元节点可以自动生长和剪除,无需在训练前预先确定自组织神经网络结构。DBTSONN1算法采用单路径自组织树中搜索最匹配叶节点(获胜神经元),DBTSONN2算法考虑了获胜神经元节点所在自组织二叉树的层次,采用双向搜索获胜叶节点,提高了搜索效率。实验结果表明,该算法在向量量化器设计方面具有很好的效果。
The advantages and disadvantages of various improved self-organizing neural network algorithms were discussed in the paper, and an Improved Dynamical Binary-tree Based Self-Organizing Neural Network （DBTSONN） was designed and implemented in detail. In the binary-tree, neuron nodes can be growing and pruning, and the self-organlzing mapping structure is flexible, not needed to be determined in advance. DBTSONNI algorithm uses single path to search the winning leaf nodes, and DBTSONN2 algorithm uses double path search, considering the hierarchical position of the winning node, which can improve the searching efficiency. The experimental results show that DBTSONN algorithm is very useful for vector quantization.
出处
《计算机应用》
CSCD
北大核心
2007年第9期2262-2266,2297,共6页
journal of Computer Applications
基金
国家自然科学基金资助项目(60602052)
福建省重点科技项目(2005H086)
关键词
自组织神经网络
动态二叉树
双向搜索机制
算法实验
self-organlzing neural network
dynamic binary-tree
double path search
algorithm experiment
