Maintaining the glucose concentration in normoglycemic range in Type I diabetic pa-tients is challenging. In this study Hoc control is applied for the insulin delivery to prevent the hyperglycaemic levels in a type I diabetic patient. A nonlinear model is linearized around nominal condition and reduced for control synthesis. Hoc controller was compared with the two other types of controller and performances shows evaluatory results.

View full abstract

Interpersonal motions and their expression using a fuzzy inference based on a mental distance type pleasure-arousal space are proposed.for a mobile eye robot in a mascot robot system. An interpersonal motion that consists of an interpersonal distance, a motion speed, and a motion trajectory is determined according to verbal information received from human in an interactive situation using a proposed fu-_y inference based on a mental distance type pleasure-arousal space. Interactive experiments with two scenarios are performed in an information recommendation situation with the mascot robot system. Subjective estimations using psychological scale and impression estimations using SD method with the factor analysis are conducted for 11 subjects. Since the results of the subjective estimation show 3.08 and 2.62 (out of 6), the validity of the fuy interpersonal motions expression is confirmed. The results of the impression estimation show that the mobile eye robot gives a feeling of closeness. The system provides user friendly and casual information recommendation, which is essential for wide spread family use.

View full abstract

The single input rule modules connected fuzzy inference method (SIRMs method) is pre-sented by Yubazaki et al. which provide same number of fuzzy rules modules as input variables. Af-terward, Seki et al. has proposed the functional-type SIRMs method (F-SIRMs method) by replacing a constant term of the consequent part of the SIRMs method with a function. This study aims to find the way to improve effectiveness of the F-SIRMs method and apply to use in medical area. The techniques in this study include multi-layer perceptrons with back propagation learning method (MLP with BP), F-SIRMs method, the generalized neural network type SIRMs method (G-NN-SIRMs method). All techniques are produced to diagnosis of the liver ailment and diabetes. The data are divided into 2 parts i.e., for training and testing and run 10 simulations. Their results are compared in order to give the lower mean square error and higher accuracy. Judging from the results of experiment, the new proposed technique gives the higher performance than conventional methods.

View full abstract

Parallel-type neuron network (PNN) is researched to improve on decreases in capabili-ties of a neuron network by the interference of learning caused between outputs of two or more outputs BP network (BPN) and the difficulty for common achievement of middle layers used.for its each output. Research to compare prediction accuracies of nonlinear time series signals prediction systems using BPN and PNN has been performed so far. However, it has not attained demonstrating an existence of dominance concerning all prediction accuracies of PNN to BPN. Then, experimental evaluation for the dominance concerning all outputs of PNN which exists theoretically by results of comparison concerning learning rules of BPN and PNN was performed using nonlinear time series signals prediction systems in this research. As a result, the dominance was shown.

View full abstract

Recently it becomes reasonable to stream video contents according to the broadband popularization. In administrative organizations the importance of free access policy to self-governing body information, such as a broadcast of the assembly, enrich their WEB pages, etc., has been widely recognized. This paper presents a support system based on movie and sound processing for building streaming media contents about the assembly information. The content includes video con-tents of the assembly and superimposed dialogue of speech text. We apply combined method of frame difference with restricted area in movie processing and discrimination based on the cepstrum in sound processing to build such contents in the support system. Through numerical examples, we show concepts of the support system

View full abstract

Though various contents are provided through the internet recently, it is not easy to find favorite contents among huge amounts of contents in terms of user's preference. In this paper, we focus on the collaborative filtering algorithm in the recommender system. We propose a fuzzy modeling approach for preference similarity model in collaborative filtering. In our approach, valid simplified fuzzy reasoning model is constructed through optimization of MAE(Mean Absolute Error). The model decides the weight of preference similarity from the value of correlation coefficient and the number of items. Through numerical experiments compared with conventional correlation coefficient based approach using Movie Lens data, the approach is found to be promising for improvement of collaborative filtering model accuracy.

View full abstract

In order to develop a data mining system for huge database mainly composed of numerical attributes, there exists necessary process to decide valid quantization of the numerical attributes. Though the clustering algorithm can provide useful information for the quantization problem, it is difficult to formulate appropriate clusters for rule extraction in terms of appropriate dimension, cluster size, and shape. In this paper, we propose a new method of quantitative association rules extraction that can quantize the attribute by applying clustering algorithm and extract rules simultaneously. From the results of numerical experiments using benchmark data, the method is found to be effective for actual applications.

View full abstract

This paper presents fast algorithms for extracting fuzzy association rules from database. The objective of the algorithm is to reduce the extracted redundant rules for the actual application, in order to improve the computational efficiency of fuzzy association rules mining. In this paper, for extracting fuzy association rules, it is assumed that the consequent part of the fuzzy rule is speci-fied in advance, i. e. before starting mining computation. This assumption corresponds to actual problems, e.g. diagnostics problem of the process, quality control action of manufacturing, and so on. The algorithm is based on the Apriori algorithm for rule extraction of the specified output field or the output fuzzy set. From the results of numerical experiments, the algorithm is found to be ef-fective compared with the conventional method in terms of computational time and redundant rule pruning.

View full abstract

In this paper, a novel method is proposed to detect faces based on unit-linking PCNN time signature and skin color segmentation, in which no training is needed. A test image is first divided into overlapped blocks and extracted PCNN time signature as the detection features, which a two-dimensional image is projected to a one-dimensional feature space. The test blocks are matched to a face template based on Euclidean distance threshold and Skin color segmentation is used to reduce the search areas and to speed up the detection procedure. The candidate face blocks are clustered to determine the face number included in a test image and the size of the face areas. The method demonstrates successful face detection over a wide range of facial variations in scale, resolution, facial expressions, in the presence of various illumina-tion conditions and under complex background from outdoor photo collections. The simulation result proves that the method is translation, rotation and scale invariant and is insensitive to facial changes.

View full abstract

In this paper, we present a new method of clustering binary data based on the combination of indiscernibility and its indiscernibility level. As a motivation of this method we con-sider core concept of classical rough sets are clustering similarities and dissimilarities of objects based on the notions of indiscernibility and discernibility. The indiscernibility level quantifies the indiscernibility of pairs of objects among other objects in information systems. The result of this paper show the dual notions of indiscernibility and its indiscernibility level play an important role in clustering information systems.

View full abstract

This paper presents a supervised higher order polynomial neural network which is called Dynamic Ridge Polynomial Neural Network. The network combines the characteristics of higher order and recurrent neural networks. It functionally extends the input space into a higher dimensional space, where linear separability is possible, without suffering from the combinatorial explosion in the number of weights. Furthermore, the presence of the recurrent link expands the network's ability for attractor dynamics and storing information for later use. In order predict the fu-ture trends of the S&P 500 signals, a Real Time Recurrent Learning algorithm was employed in training the network. Extensive simulations for the prediction of five steps ahead were performed on the signals. Experimental results indicate that the Dynamic Ridge Polynomial Neural Network demonstrated advantages in capturing chaotic movement in the signals with an improvement in the profit return, and rapid convergence over the widely known Multilayer Perceptrons.

View full abstract

Supply chain and industrial cluster are the results of competition economy. Clustering analysis enable more efficient supply chain management practices. In this paper, we focus our dis-cussion on the rough set theory for clustering supplier chain management. We propose ROSMAN (ROugh Set approach for clustering Supplier chain MANagement), an alternative algorithm for clustering supplier base management based on rough set theory taking into account maximal at-tributes dependencies in an information system. Experimental result on a supplier data set shows that ROSMAN technique is better with the baseline supplier base management clustering algorithm with respect to computational complexity and clusters purity.

View full abstract

Data classification is a vital task in large scale data mining application. DNA sequences are the basis of life and they encode all the necessary information needed to reproduce life. The size of public DNA sequence databases are growing doubling every year. This situation makes automatic classification and reduction of DNA sequences has become important for effective sequence similarity search problem. A challenge in DNA sequence similarity search is that the sequence record structure does not have any attribute that can be used for implementing classification process. In this paper, by means of filtering process an automaton based exact string matching is employed to generate a special attribute used for DNA sequence database classification and reduction. Rough sets theory provides an indiscernibility relation technique which can be used to classify and reduct the database based on some definition of 'equivalence'. The generated attribute is used.for constructing indiscernibility relation among sequences. With computational implementation, the experiments are executed to investigate the effectiveness of rough sets theory on generating DNA sequence database classification and reduction. Moreover, the experiments will demonstrate that the DNA sequence similarity search performance is significantly improved by using this approach.

View full abstract

Most of the gradient based optimization algorithms employed during training process of back propagation networks use negative gradient of error as a gradient based search direction. This paper presents a novel approach to improve the training efficiency of back propagation neural network algorithms by adaptively modifying the gradient based search direction. The proposed algorithm uses the value of gain parameter in the activation function to modify the gradient based search direction. It has been shown that this modification can significantly enhance the computational efficiency of training process. The proposed algorithm is generic and can be implemented in almost all gradient based optimization processes. The robustness of the proposed algorithm is shown by comparing convergence rates for gradient descent, conjugate gradient and quasi-Newton methods on many benchmark examples.

View full abstract

Grouping web transactions into clusters is important in order to obtain better understanding of user's behavior. Currently, the rough approximation-based clustering technique has been used to group web transactions into clusters. However, the processing time is still an issue due to the high complexity for finding the similarity of upper approximations of a transaction which used to merge between two or more clusters. On, the other hand, the problem of more than one transaction under given threshold is not addressed. In this paper, we propose RoCeT model for grouping web transactions using rough set theory. It is based on the two similarity classes which are nonvoid intersection.

View full abstract