基于EM算法的有监督LVQ神经网络及其应用

针对监督LVQ神经网络存在神经元未被充分利用以及输入样本和竞争单元之间的信息被浪费等问题。通过将EM算法引入到LVQ神经网络中,提出了基于EM聚类算法的有监督LVQ神经网络(即EMLVQ网络),从而弥补了LVQ神经网络的不足且具有EM算法提取样本信息精确的优点。通过参数的简化可以得到EMLVQ算法是软竞争格式(SCS)的一种推广。最后将它们应用于说话人辨识。实验表明,EMLVQ神经网络辨识说话人取得了很好的效果。

Supervised LVQ neural network based on EM algorithm and its application

Neurons can not be taken to full advantage and the information between feature vectors and neurons is wasted among supervised learning vector quantization (LVQ)neural networks. By introducing EM algorithm to LVQ neural network, an EMLVQ neural network which overcomes the disadvantage of LVQ neural network and possesses the merit of EM algorithm is obtainced. With parameter reduced, it is concluded that EMLVQ algorithm is a generalization of soft-competition scheme(SCS). In the end, the EMLVQ neural network is applied to speaker identification, by which its efficiency is illustrated.