Abstract

This paper proposes a new framework for training radial basis function neural networks (RBFNN). Determination of the centers of the Gaussian functions in the hidden layer of RBF neural network highly affects the performance of the network. This paper presents a novel radial basis function using fuzzy C-means clustering algorithm based on Atanassov’s intuitionistic fuzzy set (A-IFS) theory. The A-IFS theory takes into account another uncertainty parameter which is the hesitation degree that arises while defining the membership function and therefore, the cluster centers converge to more desirable locations than the cluster centers obtained using traditional fuzzy C-means algorithm. Furthermore, we make use of a new objective function obtained by Atanassov’s intuitionistic fuzzy entropy. This objective function is incorporated in the traditional fuzzy C-means clustering algorithm to maximize the good points in the class. The proposed method is used to improve the functionality of the Optimum Steepest Descent (OSD) learning algorithm. Adjusting RBF units in the network with great accuracy will result in better performance in fewer train iterations, which is essential when fast retraining of the network is needed, especially in the real-time systems. We compare the proposed Atanassov’s intuitionistic radial basis function neural network (A-IRBFNN) with fuzzy C-mean radial basis function neural network (FCMRBFNN) while both methods use OSD learning algorithm. Furthermore, the proposed A-IRBFNN is compared with other powerful fuzzy-based radial basis function neural network. Experimental results on Proben1 dataset and demonstrate the superiority of the proposed A-IRBFNN.

中文翻译：

---

基于Atanassov直觉模糊集理论的新型神经网络分类器

摘要

本文提出了一种训练径向基函数神经网络（RBFNN）的新框架。在RBF神经网络的隐藏层中确定高斯函数的中心会严重影响网络的性能。本文基于Atanassov的直觉模糊集（A-IFS）理论，提出了一种基于模糊C均值聚类算法的径向基函数。A-IFS理论考虑了另一个不确定性参数，即在定义隶属函数时出现的犹豫程度，因此，与使用传统模糊C均值算法获得的聚类中心相比，聚类中心收敛到了更理想的位置。此外，我们利用了由Atanassov的直觉模糊熵获得的新目标函数。该目标函数被合并到传统的模糊C均值聚类算法中，以最大化该类中的优点。所提出的方法用于改进最佳最速下降（OSD）学习算法的功能。高度精确地调整网络中的RBF单元将以较少的火车迭代次数获得更好的性能，这在需要快速重新训练网络（尤其是在实时系统中）时至关重要。我们将提出的Atanassov直觉径向基函数神经网络（A-IRBFNN）与模糊C均值径向基函数神经网络（FCMRBFNN）进行了比较，而两种方法都使用OSD学习算法。此外，将提出的A-IRBFNN与其他强大的基于模糊的径向基函数神经网络进行了比较。