摘要
将脉冲神经网络的高效处理能力与自组织映射神经网络相结合,构造了一种基于突触可塑性(STDP)规则的SOM脉冲神经网络模型.该网络将输入和权值用脉冲发放时间编码,符合生物信息处理机制.用STDP规则调整权值,不需要通过学习率控制收敛速度,缩短网络训练时间.使用欧氏距离的平方计算权值和样本之间的相似度,与欧氏距离法相比简化了计算,便于硬件实现.基于MATLAB仿真平台,用该网络对UCI机器学习数据库中Iris数据集进行聚类后精度达到93.33%,比传统的SOM、K-means等聚类方法更具有优越性.
The features that process signal self-organizing of cerebral cortex can be simulated by SOM neural network.Spiking neural network is a technical with best bionic performance at present,what has become one of the popular research in neural network field is that combine SOM with spiking neural network.Combined with efficient processing capabilities of spiking neural networks with SOM neural network,an improved SOM spiking neural network model based on STDP learning rule was constructed.First of all the accurate times of fired spikes were used to represent sample and weights in the network,which was in line with biological information processing mechanisms;secondly STDP learning rule was based on to adjust the weights without control convergence rate by decreasing the learning rate,which could shorten time of training network;finally the square of Euclidean distance was used to calculate the similarity of spike sequences between sample and weights,which can simplify the calculation compared with Euclidean distance method.Based on MATLAB simulation platform,the network model was used for cluster analysis of Iris dataset in UCI machine learning library,the clustering accuracy of 93.33% was gotten after training network.What was proved is that the current method has better performance compared with traditional SOM,K-means.
出处
《东北师大学报（自然科学版）》
CAS
CSCD
北大核心
2017年第3期52-56,共5页
Journal of Northeast Normal University(Natural Science Edition)
基金
国家自然科学基金资助项目(21227008)
吉林省科技发展计划项目(20130102028JC)
