A neurodynamics-based nonnegative matrix factorization approach based on discrete-time projection neural network
A neurodynamics-based nonnegative matrix factorization approach based on discrete-time projection neural network
复制标题
DOI：
10.1007/s12652-019-01550-5
复制
发表时间：
2019-10
期刊：
影响因子：
--
通讯作者：
N. Zhang;Keenan Leatham
中科院分区：
文献类型：
--
作者：
N. Zhang;Keenan Leatham
本文研究了各种NMF算法对各个分类器分类精度的影响，并对各个分类器进行了比较。研究了一种基于离散时间投影神经网络（DTPNN）的快速非负矩阵分解（NMF）算法。将NMF算法与三种分类器相结合，研究了NMF算法降维对分类器准确率的影响。两个流行的目标函数，Frobenius范数和Kullback-Leibler（K-L）的分歧不同的NMF为基础的算法在广泛的数据集的收敛目标函数值的证明。CPU运行时间方面的这些目标函数的NMF算法和数据集的不同组合也示出。此外，不同的NMF方法的收敛性进行了说明。为了测试其有效性的分类精度，三个著名的分类器的性能进行了研究和NMF算法的准确性的影响进行了评估。此外，混淆矩阵模块已被纳入到算法中，以提供额外的分类精度比较。
This paper contributes to study the influence of various NMF algorithms on the classification accuracy of each classifier as well as to compare the classifiers among themselves. We focus on a fast nonnegative matrix factorization (NMF) algorithm based on discrete-time projection neural network (DTPNN). The NMF algorithm is combined with three classifiers in order to find out the influence of dimensionality reduction performed by the NMF algorithm on the accuracy rate of the classifiers. The convergent objective function values in terms of two popular objective functions, Frobenius norm and Kullback–Leibler (K-L) divergence for different NMF based algorithms on a wide range of data sets are demonstrated. The CPU running time in terms of these objective functions on different combination of NMF algorithms and data sets are also shown. Moreover, the convergent behaviors of different NMF methods are illustrated. In order to test its effectiveness on classification accuracy, a performance study of three well-known classifiers is carried out and the influence of the NMF algorithm on the accuracy is evaluated. Furthermore, the confusion matrix module has been incorporated into the algorithms to provide additional classification accuracy comparison.