Narrow Results

This paper defines the truncated normalized max product operation for the transformation ofstates of a network and provides a method for solving a set of equations based on this operation. The operation serves as the transformation for the set of fully connected units in a recurrent network that otherwise might consist of linear threshold units. Component values of the state vector and ouputs of the units take on the values in the set {0, 0.1, …, 0.9, 1}. The result is a much larger state space given a particular number of units and size of connection matrix than for a network based on threshold units. Since the operation defined here can form the basis of transformations in a recurrent network with a finite number of states, fixed points or cycles are possible and the network based on this operation for transformations can be used as an associative memory or pattern classifier with fixed points taking on the role of prototypes. Discrete fully recurrent networks have proven themselves to be very useful as associative memories and as classifiers. However they are often based on units that have binary states. The effect of this is that the data to be processed consisting of vectors in ℜⁿ have to be converted to vectors in {0, 1}^m with m much larger than n since binary encoding based on positional notation is not feasible. This implies a large increase in the number of components. The effect can be lessened by allowing more states for each unit in our network. The network proposed demonstrates those properties that are desirable in an associative memory very well as the simulations show.

This paper, presents a novel identification approach using fuzzy neural networks. It focuses on structure and parameters uncertainties which have been widely explored in the literatures. The main contribution of this paper is that an integrated analytic framework is proposed for automated structure selection and parameter identification. A kernel smoothing technique is used to generate a model structure automatically in a fixed time interval. To cope with structural change, a hysteresis strategy is proposed to guarantee finite times switching and desired performance.

In this paper, a class of fuzzy cellular neural networks (FCNNs) with distributed delays and time-varying coefficients is studied. By employing the mathematical analysis method, some sufficient conditions are derived for ensuring that all solutions of such FCNNs converge exponentially to zero. Both the Lipschitz continuous condition on activation functions and the differentiability of time-varying delays are not required in this study.

The focus of this paper is on an attempt towards a unified formalism to manage both symbolic and numerical information based on high-level fuzzy Petri nets (HLFPN). Fuzzy functions, fuzzy reasoning, and fuzzy neural networks are integrated in HLFPN In HLFPN model, a fuzzy place carries information to describe the fuzzy variable and the fuzzy set of a fuzzy condition. An arc is labeled with a fuzzy weight to represent the strength of connection between places and transitions. A fuzzy set and a fuzzy truth-value are attached to an uncertain fuzzy token to model imprecision and uncertainty. We have identified six types of uncertain transition: calculation transitions to compute functions with uncertain fuzzy inputs; inference transitions to perform fuzzy reasoning; neuron transitions to execute computations in neural networks; duplication transitions to duplicate an uncertain fuzzy token to several tokens carrying the same fuzzy sets and fuzzy truth values; aggregation transitions to combine several uncertain fuzzy tokens with the same fuzzy variable; and aggregation-duplication transitions to amalgamate aggregation transitions and duplication transitions. To guide the computation inside the HLFPN, an algorithm is developed and an example is used to illustrate the proposed approach.

The use of intelligent models may be slow because of the number of samples involved in the problem. The identification of pulsars (stars that emit Earth-catchable signals) involves collecting thousands of signals by professionals of astronomy and their identification may be hampered by the nature of the problem, which requires many dimensions and samples to be analyzed. This paper proposes the use of hybrid models based on concepts of regularized fuzzy neural networks that use the representativeness of input data to define the groupings that make up the neurons of the initial layers of the model. The andneurons are used to aggregate the neurons of the first layer and can create fuzzy rules. The training uses fast extreme learning machine concepts to generate the weights of neurons that use robust activation functions to perform pattern classification. To solve large-scale problems involving the nature of pulsar detection problems, the model proposes a fast and highly accurate approach to address complex issues. In the execution of the tests with the proposed model, experiments were conducted explanation in two databases of pulsars, and the results prove the viability of the fast and interpretable approach in identifying such involved stars.

An iterative method of solving a set of equations based on the truncated normalized max product is described. The operation may serve as the transformation for the set of fully connected units in a fully recurrent network that might otherwise consist of linear threshold units. Because of truncation and normalization the network acting under this transformation has a finite number of states and components of the state vector are bounded. Component values however are not restricted to binary values as would be the case if the network consisted of linear threshold units but can now take on the values in the set {0, 0.1,..0.9, 1}. This means that each unit although still having discrete output can provide finer granularity compared to the case where a linear threshold unit is used. Truncation is natural in hardware implementation where only a finite number of places behind the decimal are retained.

We introduce and study a new concept of fuzzy computing units. This construct is is aimed at coping with "negative" (inhibitory) information and accommodating it in the language of fuzzy sets. The essential concept developed in this study deals with computing units exploiting the concept of balanced fuzzy sets. We recall how the membership notion of fuzzy sets can be extended to the [-1,1] range giving rise to balanced fuzzy sets and then summarize properties of augmented (extended) logic operations for these constructs. We show that this idea is particularly appealing in neurocomputing as the "negative" information captured through balanced fuzzy sets exhibits a straightforward correspondence with inhibitory processing mechanisms encountered in neural networks. This gives rise to interesting properties of balanced computing units when compared with fuzzy and logic neurons developed on the basis of classical logic and classical fuzzy sets. Illustrative examples concerning topologies and properties and learning of balanced fuzzy computing units are included. A number of illustrative examples concerning topologies, properties and learning of balanced fuzzy fuzzy computing units are included.

In this study, we proposed an alternative operation of fuzzy arithmetic on L-R fuzzy numbers by three parameters of mode, left spread and right spread. Then, based on this approximation method, a new learning algorithm of a fully fuzzified neural network was developed in which the L-R fuzzy numbers were considered as the fuzzy signals. While the forward operations of fuzzy signals were based on the proposed three-parameter fuzzy arithmetic approximation method, the backward learning adopted a back-propagation learning procedure with a measurable error function. The learning algorithm was illustrated by an example of the recognition of fuzzy IF-THEN rules. The simulation result showed that the proposed approximation method used in such learning model was efficient and accurate.

Some neural network related methods have been applied to nonlinear fuzzy regression analysis by several investigators. The performance of these methods will significantly worsen when the outliers exist in the training data set. In this paper, we propose a training algorithm for fuzzy neural networks with general fuzzy number weights, biases, inputs and outputs for computation of nonlinear fuzzy regression models. First, we define a cost function that is based on the concept of possibility of fuzzy equality between the fuzzy output of fuzzy neural network and the corresponding fuzzy target. Next, a training algorithm is derived from the cost function in a similar manner as the back-propagation algorithm. Last, we examine the ability of our approach by computer simulations on numerical examples. Simulation results show that the proposed algorithm is able to reduce the outlier effects.

In this paper, we first examine limitations of fuzzy neural networks. We find the following. (1) If training errors are the main concerns, Spline can perform better than the generalized dynamic fuzzy neural network (GD-FNN). (2) If the model is nonlinear with a disturbance term, the testing error of the GD-FNN is very large. If the model is chaotic with a disturbance term, both the training error and testing error of the GD-FNN are very large. (3) Using a sequential algorithm as in the GD-FNN, we would always be trapped at the local minima rather than the global minimum. In addition, we propose to use the characteristics among moments and fuzzy rules to identify the density function in advance.

This work presents an automated method of segmentation of faces in color images with complex backgrounds. Segmentation of the face from the background in an image is performed by using face color feature information. Skin regions are determined by sampling the skin colors of the face in a Hue Saturation Value (HSV) color model, and then training a fuzzy min-max neural network (FMMNN) to automatically segment these skin colors. This work appears to be the first application of Simpson's FMMNN algorithm to the problem of face segmentation. Results on several test cases showed recognition rates of both face and background pixels to be above 93%, except for the case of a small face embedded in a large background. Suggestions for dealing with this difficult case are proffered. The image pixel classifier is linear of order O(Nh) where N is the number of pixels in the image and h is the number of fuzzy hyperbox sets determined by training the FMMNN.

There have been many applications in the area of handwritten character recognition. Over the last decade much research has gone into developing algorithms to accurately convert captured images of handwriting to text. At the same time, research into neuro fuzzy classification models has proven to solve many complex problems. In this paper, Adaptive Neuro Fuzzy Inference System (ANFIS) and Evolving Fuzzy Neural Network (EFuNN) was investigated and studied in detail on how these two models can be used to perform handwritten digits classification. Results of the experiments show great potential of the EFuNN over the ANFIS for practical implementation of the handwritten digit recognition.

In this study, a new instance selection method that combines the neural network with weighted fuzzy memberships (NEWFM) and Takagi–Sugeno (T–S) fuzzy model was proposed to improve the classification accuracy of healthy people and Parkinson’s disease (PD) patients. In order to evaluate the proposed instance selection for the classification accuracy of healthy people and PD patients, foot pressure data were collected from healthy people and PD patients as experimental data. This study uses wavelet transforms (WTs) to remove the noise from the foot pressure data in preprocessing step. The proposed instance selection method is an algorithm that selects instances using both weighted mean defuzzification (WMD) in the T–S fuzzy model and the confidence interval of a normal distribution used in statistics. The classification accuracy was compared before and after instance selection was applied to prove the superiority of instance selection. Classification accuracy before and after instance selection was 77.33% and 78.19%, respectively. The classification accuracy after instance selection exhibited a higher classification accuracy than that before instance selection by 0.86%. Further, McNemar’s test, which is used in statistics, was employed to show the difference in classification accuracy before and after instance selection was applied. The results of the McNemar’s test revealed that the probability of significance was smaller than 0.05, which reaffirmed that the classification accuracy was better when instance selection was applied than when instance selection was not applied. NEWFM includes the bounded sum of weighted fuzzy memberships (BSWFMs) that can easily show the differences in the graphically distinct characteristics between healthy people and PD patients. This study proposes new technique that NEWFM can detect PD patients from foot pressure data by the BSWFMs embedded in devices or systems.

We present a new approach to acquisition of comprehensible fuzzy rules for fuzzy modeling from data using Evolutionary Programming (EP). For accuracy of model, it is effective to allow overlapping of membership functions with each other in the fuzzy model. From the viewpoint of knowledge acquisition, it is desirable that the model has a smaller number of membership functions with less overlapping. Considering the trade-off between the precision and the clarity of the fuzzy model, this paper presents an acquisition method of comprehensible fuzzy rules form the identified model that satisfies the desired accuracy. The approach clearly distinguishes modeling phase and re-evaluation phase. The accurate model of unknown system in the modeling phase is to be obtained by, for example, fuzzy neural network (FNN) such as a radial basis function network, using EP. The simplified model in the re-evaluation phase can mainly be used for knowledge acquisition from unknown system. A numerical experiment was done to show the feasibility of the proposed algorithm.

This paper introduces some basic related theories of data mining technique, including the basic perception and main related methods. Then this paper presents the disadvantages of the conventional methods and fuzzy neural networks, which has the advantages of high precision and fast convergence speed. Moreover, the urban traffic incident forecast system has been developed which takes advantages of this improved method. Finally, this paper discusses the promising developments and some related applications of the information mining technique based on this method.