Learning Algorithm with Gaussian Membership Function for Fuzzy RBF Neural Networks

In this paper a new learning algorithm for Fuzzy Radial Basis Function Neural Networks is presented, which is characterized by its fully-unsupervising, self-organizing and fuzzy properties, with an associated computational cost that is fewer than other algorithms. It is intended for pattern classification tasks, and is capable of automatically configuring the Fuzzy RBF network. The methodology shown here is based on the self-determination of network architecture and the self-recruitment of nodes with a gaussian type of activation function, i.e. the center and covariance matrices of the activation functions together with the number of tuned and output nodes. This approach consists in a mix of the “Thresholding in Features Spaces” techniques and the updating strategies of the “Fuzzy Kohonen Clustering Networks” introducing a Gaussian Membership function. Its properties are the same as those of the traditional membership function used in Fuzzy c-Means clustering algorithms, but with the membership function proposed here it lets a nearer relationship exist between learning algorithm and network architecture. Data from a real image and the results given by the algorithm are used to illustrate this method.

[1]  Eduardo Bayro-Corrochano,et al.  Self-organizing neural-network-based pattern clustering method with fuzzy outputs , 1994, Pattern Recognit..

[2]  James C. Bezdek,et al.  Pattern Recognition with Fuzzy Objective Function Algorithms , 1981, Advanced Applications in Pattern Recognition.

[3]  Sukhan Lee,et al.  A Gaussian potential function network with hierarchically self-organizing learning , 1991, Neural Networks.

[4]  Shang-Liang Chen,et al.  Orthogonal least squares learning algorithm for radial basis function networks , 1991, IEEE Trans. Neural Networks.

[5]  Mohamad T. Musavi,et al.  On the training of radial basis function classifiers , 1992, Neural Networks.

[6]  Sang Uk Lee,et al.  On the color image segmentation algorithm based on the thresholding and the fuzzy c-means techniques , 1990, Pattern Recognit..

[7]  J.C. Bezdek,et al.  Computing with uncertainty , 1992, IEEE Communications Magazine.

[8]  Edward Roy Davies Design of optimal gaussian operators in small neighbourhoods , 1987, Image Vis. Comput..

[9]  J. Carrabina,et al.  Neural-like network model for color images analysis systems , 1994, Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94).

[10]  John Moody,et al.  Fast Learning in Networks of Locally-Tuned Processing Units , 1989, Neural Computation.

[11]  Moustafa M. Fahmy,et al.  Probabilistic Winner-Take-All Learning Algorithm for Radial-Basis-Function Neural Classifiers , 1994, Neural Computation.

[12]  James C. Bezdek,et al.  Fuzzy Kohonen clustering networks , 1994, Pattern Recognit..

[13]  D.R. Hush,et al.  Progress in supervised neural networks , 1993, IEEE Signal Processing Magazine.

[14]  Anders Hald,et al.  Statistical tables and formulas , 1952 .