Ying-Kuei Yang

Robust neural-fuzzy method for function approximation

The back propagation (BP) algorithm for function approximation is multi-layer feed-forward perceptions to learn parameters from sampling data. The BP algorithm uses the least squares method to obtain a set of weights minimizing the object function. One of main issues on the BP algorithm is to deal with data sets having variety of data distributions and bound with noises and outliers. In this paper, in order to overcome the problems of function approximation for a nonlinear system with noise and outliers, a robust fuzzy clustering method is proposed to greatly mitigate the influence of noise and outliers and then a fuzzy-based data sifter (FDS) is used to partition the nonlinear systempsilas domain into several piecewise linear subspaces to be represented by neural networks. Two experiments are illustrated and these results have shown that the proposed approach has good performance in various kinds of data domains with data noise and outliers.

A Robust Fuzzy Clustering Approach And Its Application To Principal Component Analysis

Abstract A robust fuzzy clustering approach is proposed to simplify the task of principal component analysis (PCA) by reducing the data complexity of an image. This approach performs well on function curves and character images that not only have loops, shazp corners and intersections but also include data with noise and outliers. The proposed approach is composed of two phases: fustly, input data are clustered using the proposed distance analysis to get good and reasonable number of clusters; secondly, the input data are further re-clustered by the proposed robust fuzzy c-means (RFCM) to mitigate the influence of noise and outlier data so that a good result of principal components can be found. Experimental results have shown the approach works well on PCA for both curves and images despite their input data sets include loops, corners, intersections, noise and outliers.

A ROBUST APPROACH OF FUZZY CLUSTERING ON DATA SETS WITH NOISE AND OUTLIERS

In this paper, a new robust fuzzy clustering approach is proposed for better performing principal component analysis (PCA) on function curves and character images that not only have loops, sharp corners, and intersections but also are bound of noise and outlier data. The proposed method is composed of two phases: firstly, input data are clustered using the proposed distance analysis to get good initial cluster centers and a reasonable number of clusters; secondly, the input data are further reclustered by the proposed robust fuzzy c-means (RFCM) based on the results obtained in the first phase to overcome the influence of noise and outlier data so that a good result of principal components can be found. Several function curves and Chinese character images are given to illustrate the effectiveness of the proposed method. Experimental results have demonstrated that the proposed approach works very well on PCA for both curves and images despite the fact that their input data sets may include loops, corners, intersections, noise, and outlier information.

Study of nonlinear phenomena and chaotic control in current-programmed forward converters

It is evident that effective control over chaos in the forward converter is very important for engineering applications. In this paper, a mathematical model of the forward converter is firstly derived. Secondly, the nonlinear phenomenon in the forward converter is discussed, and then we apply the method of time-delay feedback control over chaos in the forward converter. The validity of this method is confirmed by simulation results.

A novel learning framework of CMAC via grey-area-time credit apportionment and grey learning rate

The advantages of CMAC neural network are fast learning convergence, capable of mapping nonlinear functions quickly due to its local generalization of weight updating, simple architecture, easily processing and hardware implementation. In the training phase, the disadvantage of some CMAC models with a larger fixed learning rate is the unstable phenomenon. The smaller learning rate would cause slower convergence speed. In the aspect, we propose grey learning rate for training phase. We incorporate the grey relational analysis with the number of training iterations to get an adequate learning rate for better convergence performance. In addition, a serious problem of learning interference reduces learning speed and accuracy. The idea is that the error correcting must be proportional to the inverse of learning times, trained input area and grey relational grade for the addressed hyper cube. A credit apportionment adopts the idea to provide fast and accurate learning effects. This paper proposes a novel learning framework of CMAC for better performance and real-time applications. From the simulation results, it is evident that the proposed algorithm provides more accurate and fast convergence in the early cycles of training phase and also becomes significant in the follow-up cycles.

A novel consumer clock device based on grey relational analysis

People have the need to use a novel consumer clock device for home automation, personal security and convenience. This paper presents the design and implementation of the innovative system. The system architecture which is controlled by Micro-Controller Unit (MCU) includes main alarm clock module, lucky module and security module. In the main alarm clock module, we create a novel hourglass clock model and intelligent alarm clock features based on grey relational analysis. In the lucky module, we design a unique structure of live round according to the medical theory of psychology and hashing function for randomization. In the live round, it includes the fortune of love, career, family, and health. It adopts funny and encouraged voice to transmit messages. The system also provides the function of security using infrared sensors and third-generation (3G) mobile phone unit. The system has been successful and gets the award in Taiwan.

Function approximation using robust fuzzy-GreyCMAC method

In this paper, we propose a novel GreyCMAC with robust FCM (RFCM) method for function approximation. The advantages of CMAC neural network are fast learning convergence, capable of mapping nonlinear functions quickly due to its local generalization of weight updating. In order to overcome the problems of function approximation for a nonlinear system with noise and outliers, a robust fuzzy clustering method (RFCM) is proposed to effectively mitigate the influence of noise and outliers and then a GreyCMAC model is used to learn the nonlinear systems features for function approximation. There are two core ideas in the proposed method: (1) The robust fuzzy c-means algorithm (RFCM) is proposed to greatly mitigate the influence of data noise and outliers; and (2) A Grey-based CMAC (GreyCMAC) is proposed to locate a given fine piecewise linear data domain by RFCM so that a neural network can be constructed for function approximation. The conducted experimental results clearly indicate that the proposed approach provides good performance.

A Robust Fuzzy Clustering Approach and its Application to Function Approximation

Function approximation is to model a desired function or an input-output relation from a set of input-output sample data that unfortunately often suffer from noise and outliers in real systems. To overcome this problem, this paper presents an unsupervised fuzzy model construction approach to extract fuzzy rules directly from numerical input-output data for nonlinear function approximation problems with noise and outliers. There are two core ideas in the proposed method: (1) The robust fuzzy c-means (RFCM) algorithm is proposed to greatly mitigate the influence of data noise and outliers; and (2) A fuzzy-based data sifter (FDS) is proposed to locate good turning-points to partition a given nonlinear data domain into piecewise clusters so that a Takagi and Sugeno fuzzy model (TS fuzzy model) can be constructed with fewer rules. Two experiments are illustrated and their results have shown the proposed approach has good performance in various kinds of data domains with data noise and outliers.

A weight-featured and data-distribution-based fuzzy pattern classification approach

Most research proposed so far for fuzzy pattern classification has not considered the characteristic of data distribution in a given data set during the process of clustering. This paper proposes an approach that can appropriately cluster a given data set automatically based on data distribution of a given data set without the need of specifying the number of resultant clusters and setting up subjective parameters. Some special data distributions, such as stripe- or belt-shaped distributions, can therefore be nicely clustered for better pattern classification. Statistical concept is applied to define weights of pattern features so that the weight of a pattern feature is proportional to the contribution the feature can provide to the task of pattern classification. The proposed weight definition not only reduces the dimensionality of feature space so as to speed up the classification process but also increases the accuracy rate of classification result. The experiments in this paper demonstrate the proposed method has fewer fuzzy rules and better classification accuracy than other related methods.