摘要
针对人脑网络状态观测矩阵高维无特征的特点,给出了一种基于深度自动编码器(DAE)的降维算法。利用深度学习网络,将高维的人脑网络空间表达映射到低维的本质特征空间中,为进一步提炼脑网络的动态性能提供了基础。实验结果证明:应用该方法可以达到有效的降维效果,且降维后脑网络状态通过自组织特征映射聚类具有一定的规律性,从而为脑网络的动态特性研究提供了基础。
Aiming at properties of unlabelled and high-dimensional of brain network state observation matrix, a dimensional reduction algorithm based on deep autoeneoder (DAE)is presented. By using the deep learning network algorithm, high dimensional brain network character space is mapped into low dimensional fundamental feature space and so basis for refining the dynamic performance of brain network is provided. Experimental results show that this method can reduce the dimension of human brain status observation matrix effectively, and when applying self-organizing feature map clustering method to the data after dimension reduction ,the clustering results show regularity which may give us some hints for further research, provide basis for study of dynamic features of brain networks.
出处
《传感器与微系统》
CSCD
2017年第1期9-12,共4页
Transducer and Microsystem Technologies
基金
国家自然科学基金资助项目(61263017)
云南省自然科学基金资助项目(KKSY201303120)
关键词
脑功能网络
深度自动编码器
降维
自组织特征映射
无监督聚类
functional brain network
deep autoencoder ( DAE )
dimension reduction
self-organizing featuremaps
unsupervised clustering