P300 is one of the event-related potentials that arise when an infrequent stimulus appears. Oddball tasks are often used to measure P300. This research attempts to create a game-based oddball task for users to enjoy the tasks. There are several elements in a game. Score, feedback system, background music (BGM) are some of them. This paper discusses the effects of them in oddball tasks.

Recently, educational-support robots, which support learning attract many people's attentions. In previous research the robot teaches how to solve some questions only. However, it is difficult for learners to improve their applied skill and inquiring mind, because it cannot prompt learners to deliberate. Thus, this study develops a robot which support based on cognitive apprenticeship theory. The previous study of pedagogy reported that teaching based on cognitive apprenticeship theory can prompt learners to deliberate. Therefore, learner who was taught based on cognitive apprenticeship theory can be improved the applied skills and inquiring mind. In this paper, we investigates learning effect and impression effect of learners who learn with the robot which supports based on cognitive apprenticeship theory.

In recent years, educational-support robots that can assist human learners have been attracting the attention of researchers. However, learners feel that these robots have monotonous behaviors, making collaborative learning with the robot a boring experience. To solve this problem, a previous study proposed using the sympathy-expression method in which the robot expresses its own emotions autonomously on the basis of the answer time. However, we propose the notion that the writing time should also be considered. The writing time is defined as the time taken by the learners to answer the questions. When learners wrote a lot, the answer time became long and the robot (in the previous method) used to express emotions of sleep. To empathize with the learners, it is important for the robot to express emotions of arousal. Therefore, in this paper, we propose the sympathy-expression method, which expresses emotions based on the writing time.

Recently, educational-support robot have been attracting increasing attention as learning-support devices. Previous research has proposed the sympathy expression method that robot expresses emotions which sympathize with learner. In previous research, the sympathy expression method has been used for the robot which expresses its emotion by face. Sympathy expression method can be contracted based on fuzzy logic. However, no research for sympathy expression method using the robot which expresses its emotion by using body or unifying body and face. Therefore, this paper examines the learning effect and the impression about three types robots which have different emotion expression methods in long term experiment.

With the growth of robot technology, more robots are being designed to support learning. Most studies in the field have focused on robot behavior, with only a few studies focusing on robot utterances. Correspondingly, the manner in which such utterances affect learning is poorly understood. This study investigates the effects of collaborative learning wherein a robot encourages a learner. We conducted an experiment to compare the learning effect in three groups. In the first group, learning was conducted with a robot that supplied praises using onomatopoeia, i.e., the representation of an object or state based on the sounds associated with it. In the second group, a robot supplied praises using adjectives or adverbs. In the third group, a robot supplied praise without using onomatopoeia, adjectives, or adverbs. The results of this study suggest that collaborative learning using the first or third method is more effective than that involving adjective/adverb communication.

Previous studies have reported that the peak latency of P300 in healthy adults increases with age and depends on task difficulty. We study the relations between the peak latency of P300 and task difficulty and age of subjects through three types of experiments. First experiment is a visual stimulation task with a single target stimulus and one or two non-target stimuli. Second experiment is a visual stimulation task with a single circle stimulus with variable diameter and a single non-target circle stimulus with invariable diameter. Third experiment is a visual stimulation task with a variable vertex target and non-target stimuli. The experimental results show that the relation between task difficulty and the peak latency of P300 changes with age.

As robot technology grows, more attention is being paid to educational-support robots that assist in learning. For example, an educational-support robot may support students in their school life or help them learn other languages. However, users tend to lose interest in educational-support robots. To solve this problem, a model of emotional expressions has been proposed in human-agent interaction studies. This model, in which the screen agent is programmed to express emotions autonomously, has proven to be effective for interactions between agents and humans. Thus, this paper proposes an emotional expression model for use with educational-support robots. Using simulations, this paper also investigates the number of emotions that a robot can express.

Recently, more educational-support robots, which support learning, are paid attention to. However, the problem of these robots is that a user loses his/her interest in them. To solve this problem, some studies which focus on emotional expression models have been reported in Human-Agent-Interaction. The model of emotional expressions is defined as the agent expressing its emotions autonomously. Although these models have been shown to be beneficial for effective interaction between an agent and a human, no reports have addressed the educational-support robots using these models. Thus, this paper studies how much learning effect with a robot which expresses the emotion by using the model of emotional expression can be prompted for learners in a collaborative learning.

Latent Dirichlet allocation (LDA) is a topic model that has been applied to various fields, including user profiling and event summarization on Twitter.When LDA is applied to tweet collections, it generally treats all aggregated tweets of a user as a single document.Twitter-LDA, which assumes a single tweet consists of a single topic, has been proposed and has shown that it is superior in topic semantic coherence.However, Twitter-LDA is not capable of online inference.In this study, we extend Twitter-LDA in the following two ways.First, we model the generation process of tweets more accurately by estimating the ratio between topic words and general words for each user.Second, we enable it to estimate the dynamics of user interests and topic trends online based on the topic tracking model (TTM), which models consumer purchase behaviors.

In the field of marketing, a questionnaire is one of the most important approaches in order to research the market or to design a marketing strategy. On the other hand, people have a variety of individuality recently, then respondents have various impressions on evaluation objects. In the analysis of collected questionnaire data, it is important not only to analyze overall trends but also to discover minority groups which have strong impressions but are different from general groups. It is, however, difficult to extract minority groups by conventional cluster analysis applied to questionnaire data, because they generally aim at extracting majority groups or making a rough clustering. In this paper, we propose the extraction method of minority groups in questionnaire data using the spectral clustering method which considers local similarity and extracts the clusters having less connection to general groups.

The number of available items in online shops are increasing by the spread of the Internet recently. Though users have a wide range of choices, they need to find their favorite items from a huge amount of information. Thus, a variety of recommendation systems are currently in use. "Accuracy" is the most important index in these recommendation systems. However, not only "Accuracy" but also "Serendipity" is said to be needed in terms of user satisfaction recent years. In this paper, we introduce a recommendation method of collaborative filtering based on association analysis which is one of the data mining techniques. We aim to improve Serendipity keeping Accuracy high by using the evaluation information that are rated differently from a target user. In addition, we show that Accuracy and Serendipity can be adaptable by a parameter in the proposed method. This paper compares the proposed method with a conventional method in terms of the performance of Accuracy and Serendipity.

We investigated the effect of robot's utterances using onomatopoeia on collaborative learning. The robot was designed to praise or comfort by using onomatopoeia when learners are given problem to solve through a learning system. When learners can correctly solve a problem, the robot praises the learner's success. When learners cannot solve it, the robot comforts the learners to keep working at it. Eight college students learns mathematics by using a learning system with a robot for three weeks and took exams. We found that a robot could comfort learners that used onomatopoeia more than a robot that did not use onomatopoeia. This suggests that the robot that praises or comforts by using onomatopoeia helps learners maintain their motivation in collaborative learning.

P300 speller is the communication tool based on Brain Computer Interfaces (BCIs) which allow users to input letters only by thoughts. It uses P300, one of the event-related potential (ERP), as the target feature. In P300 speller, another person starts and closes the system. Therefore, a user cannot switch P300 speller ON/OFF by himself/herself. To solve this problem, an asynchronous P300 speller which can control ON/OFF based on the user's intention of input is needed. In recent years, the intention classification method with additional pre-training has been proposed. In the additional pre-training, the classifier trains non-control state data which is recorded when the user does not input. However, the additional pre-training causes another burden and usage restrictions. In this paper, we propose and study an intention classification method using only training data in which a user inputs letters and an asynchronous system in P300 speller based on the user's intention of input.

In this study, we sought to examine how the behavior of a robot can prompt collaborative learning with a human. We focus on constructive interaction that has been regarded as a foundation of collaborative learning and occurs when two learners alternately solve a question. For this, the robot is designed to alternately perform speaker and listener motions for constructive interaction with a human. With the speaker motion, the robot explains a solving method to the partner and solves a question. Moreover, the robot improves its accuracy rate as learning progresses. With the listener motion, the robot does not solve a question and instead pays attention to the partner who is solving the question. The robot learns while solving a question issued by a learning system with a human student. College students recruited as volunteers learned with learning system with the robot for one month and were videoed during that time to see how they learned. The results of examination suggest that the robot prompts learners to learn by constructive interaction in collaborative learning and possibly gains the same learning effect as collaborative learning between two humans.

In the conventional P300-based Brain Computer Interface, characters or commands are matrix-arrayed and each row or column is randomly and successively intensified. In this interface, a user needs to move his or her eye gaze to the character or the command that the user desires, which could be a problem for those with severe motor disability. To solve this problem, the use of Rapid Serial Visual Presentation (RSVP) in the P300-based BCIs has been proposed. However, it is said that a longer classification time is needed in RSVP, because characters or commands are presented one by one in this method. However, the increase of the classification time depends on the number of choices. Thus, this paper compares the performance of RSVP and the conventional interface with fewer characters or commands. Moreover, reliability based selective stimulus presentation is applied to RSVP, and this paper studies on the improvement of classification time.

Recently, the document information managed in companies tend to be more complex and various. Specification documents are used for the technical transfer and inheritance of manufactures and services. However, the description and the meaning of the component words in specifications are often inconsistent or multiple, because a specification document is made by the persons in charge of various parts. Then the readers may misunderstand the contents of specifications by them. This paper focuses on synonyms, multiple description of words for a meaning or a word, in specification documents and proposes an extraction method of them considering the co-occurrence words of component words. This paper applies the proposed method to a test data, in which some words in an actual specification document are replaced with another words, and studies the effectiveness of the proposed method.

Genetic Algorithm (GA)[1] is one of the most effective methods in the application to optimization problems. Recently, Multi-objective Genetic Algorithm (MOGA) is focused on in the engineering design field. In this field, the analysis of design variables in the acquired Pareto solutions, which gives the designers useful knowledge in the applied problem, is important as well as the acquisition of advanced solutions. This paper proposes a new visualization method using Isomap which visualizes the geometric distances of solutions in the design variable space considering their distances in the objective space. The proposed method enables a user to analyze the design variables of the acquired solutions considering their relationship in the objective space. This paper applies the proposed method to the conceptual design optimization problem of hybrid rocket engine and studies the effectiveness of the proposed method. It shows that the visualized result gives some knowledges on the features between design variables and fitness values in the acquired Pareto solutions.

A lot of companies carry out questionnaires. These questionnaires often have questions which need respondents to answer by free description. It is, however, inefficient for an analyzer to read whole texts to get outlines or classify them, or it is difficult to correctly analyze them without subjective biases. The authors have proposed "HK Graph" (Hierarchical Keyword Graph) which is a support tool for text mining. HK Graph can visualize the relationships among attributes and words with hierarchical graph structure based on frequency of co-occurrence. However, the result of HK Graph is not enough helpful for the analyzer to grasp the outlines of the texts and extract opinions from them, because it regards divided words as different ones unless they perfectly match and that makes the visualized result complicated. This paper presents a new visualization method for the HK Graph incorporating an aggregating words method based on concepts of words. An experiment is carried out by applying the proposed method to actual questionnaire data on disasters and studies the effectiveness of the proposed method.

Clustering is one of the most useful methods for understanding similarity among data. However, most conventional clustering methods do not pay sufficient attention to the geometric properties of data. Geometric algebra (GA) is a generalization of complex numbers and quaternions able to describe spatial objects and the relations between them. This paper uses conformal GA (CGA), which is a part of GA, to transform a vector in a real vector space into a vector in a CGA space and presents a proposed new clustering method using conformal vectors. In particular, this paper shows that the proposed method was able to extract the geometric clusters which could not be detected by conventional methods.

Genetic Algorithm is one of the most effective optimization algorithms, on which a lot of studies have been reported. Some studies on the application of island model, which is one of the representative methods to keep a diversity of solutions, to Multi-Objective Genetic Algorithm (MOGA) have been conducted. In MOGA, it is difficult to find the solutions which satisfy all objective functions because of their tradeoff. Especially when there are many objective functions, it is obvious that it needs a lot of time to search for effective Pareto solutions and find them. This paper proposes the interactive way of addition and deletion of islands to the original ones based on user's requirements with the visualization of acquired solutions in island model for MOGA. This paper applies the proposed method to Nurse Scheduling Problem (NSP) using the visualization by Principal Component Analysis (PCA). Through the experiment, it is confirmed that an interactive tuning of the weights for the objective functions leaded to the acquisition of better Pareto solutions which a user wants while they are difficult to be acquired by the prepared weights.

A brain-computer interface (BCI) is a system which could enable patients like those with amyotrophic lateral sclerosis to control some equipment and to communicate with other people, and has been anticipated to be achieved. One of the problems in BCI research is a trade-off between transmission speed and accuracy. In the field of data transmission, on the other hand, reliability-based hybrid automatic repeat request (RB-HARQ), one of the error control methods, has been developed to achieve both of the performances. The authors, therefore, have considered BCIs as communications between users and computers, and applied reliability-based ARQ, customized RB-HARQ, to BCIs. It has been shown that the proposed method is superior to other error control methods in two-class classification. In this paper, the proposed method is extended to deal with multi-class classification of EEG data, and is shown to be effective in multi-class problems.

Recently, companies often carry out questionnaire(s) and develop marketing strategies. There are usually two types of forms for the answer of a questionnaire. One is the form to select prepared answers and the other is free text form. The true message might be in the text form rather than the numerical part, then the analysis of free text form is needed. The amount of text in a questionnaire is, however, usually large and difficult to read whole text data for analysis. This study tries to develop a free text analysis support system which visualizes relationships among respondents based on their texts and shows their opinions using graph structure of keywords. First, this paper proposes the extraction method of important keywords in their opinions based on the modification relationships. Next, it clusters the respondents interactively on visible space using MDS. Finally, it shows their opinions using HK Graph which can visualize the relationship among words with hierarchical network structure based on the co-occurrence information for the keyword graph.

Several kinds of values have been employed with respect to the diversification of individuality in the market. Some of these values are currently supported by only a few people, who are referred to as a ldquominority grouprdquo. However, there is the possibility that such groups will grow into majority groups with changes in historical background or people's sensitivity. It is both important and effective for market analysis to determine these minority groups at an early stage. Companies often employ questionnaires to develop marketing strategies or design new products, which offer a chance to determine these minority groups. With conventional methods, respondents to a questionnaire are classified based on such attributes as gender and age, and then the classified groups are analyzed or compared. Although conventional analysis is effective for grasping the overall tendency of the evaluation data, it is difficult to determine minority groups because of the diversity of individuality. On the other hand, we have proposed clustering methods based on the tendencies of the answers to the questionnaire. This paper proposes a new method for visualizing the evaluated data based on both the obtained values and their correlation with cluster respondents interactively in the visible space. This paper applies the proposed method to Web questionnaire data and shows that an analysis of the results effectively assists us to determine minority groups.

Most conventional methods of feature extraction for pattern recognition do not pay sufficient attention to inherent geometric properties of data, even in the case where the data have spatial features. This paper introduces geometric algebra to extract invariant geometric features from spatial data given in a vector space. Geometric algebra is a multidimensional generalization of complex numbers and of quaternions, and it ables to accurately describe oriented spatial objects and relations between them. This paper proposes to combine several geometric features using Gaussian mixture models. It applies the proposed method to the classification of hand-written digits.

Most conventional methods of feature extraction do not pay much attention to the geometric properties of data, even in cases where the data have spatial features. In this study we introduce geometric algebra to undertake various kinds of feature extraction from spatial data. Geometric algebra is a generalization of complex numbers and of quaternions, and it is able to describe spatial objects and relations between them. This paper proposes to use geometric algebra to systematically extract geometric features from data given in a vector space. We show the results of classification of hand-written digits, which were classified by feature extraction with the proposed method.

This paper applies the fitness inference method to interactive genetic algorithm based on paired comparison (PC-IGA). PC-IGA enables users to reduce the mental burden for evaluation by using paired comparison. Fitness inference method can decrease evaluation times on EC by using inferred fitness value instead of actual evaluation for candidate solutions based on the information of actually evaluated solutions in the past generations. However, PC-IGA does not give enough information to apply fitness inference method. This paper investigates effective application of the fitness inference method to PC-IGA through experiments with simulated evaluation. The experimental results show that fitness inference method can decrease the number of actual evaluation times by 40% comparing with the normal PC-IGA.

Performance in searching solutions by Multi-Objective Genetic Algorithm (MOGA) depends on genetic operators and/or their parameters. For comparison of the performance with some genetic operators and/or parameters, it has been usually employed the transitions of fitness values through actual applications or the number/performance of acquired Pareto solutions in multi-optimization problems. This paper proposes a visualizing method of search process for MOGA, which can visualize relative distances among chromosomes in search process and give information of not only the performance but also the effects of the genetic operations such as the diversity of chromosomes. This method uses Self-Organizing Map (SOM) for the visualization. This paper applies Non Dominated Sorting Genetic Algorithm-II (NSGA-II) to ZDT2 and FON test functions and shows obtained nondominated solutions and visualization results. This paper also shows that the visualized data enables us to interpret the differences in search processes and to get new information to determine efficient genetic operators and their parameters.

Performance in searching solutions by Multi-Objective Genetic Algorithm (MOGA) depends on genetic operators and/or their parameters. For comparison of the performance with some genetic operators and/or parameters, it has been usually employed the transitions of fitness values through actual applications or the number/performance of acquired Pareto solutions in multi-optimization problems. This paper proposes a visualizing method of search process for MOGA, which can visualize relative distances among chromosomes in search process and give information of not only the performance but also the effects of the genetic operations such as the diversity of chromosomes. This method uses Self-Organizing Map (SOM) for the visualization. This paper applies Non Dominated Sorting Genetic Algorithm-II (NSGA-II) to FON test function and shows obtained nondominated solutions and visualization results. This paper also shows that the visualized data enables us to interpret the differences in search processes and to get new information to determine efficient genetic operators and their parameters. In addition, this paper applies the proposed method to Nurse Scheduling Problem (NSP), which is one of the complex multi-objective optimization problems in real world, and shows effective feedback to genetic operations can be done.

Interpretability of fuzzy models has become one of the major topics in the field of fuzzy modeling. Visualization that makes input-output relationships interpretable is effective in extracting useful knowledge from unknown data. This paper presents visualization method that considers the visibility of fuzzy models. This method identifies clusters that have different statistical features, and projects the data to the "fusion axes", which are linear combinations of the multiple input variables, considering the distribution of each cluster in the projected space. This paper applies the proposed method to artificial data and also to collected data from the mobile robot, and shows that the proposed method can extract useful knowledge from the obtained visible and interpretable models.

Most regression studies focus on the conditional mean estimation. A more informative description of the conditional distribution can be obtained through the conditional quantile estimation. We study on nonparametric conditional quantile estimator using support vector (SV) regression approach. We show that a slight modification of Vapnik's /spl epsi/-insensitive SV regression leads to a nonparametric conditional quantile estimator with L/sub 2/ regularization. With the great flexibility in nonparametric approach, it is quite possible that two or more estimated conditional quantile functions at different orders could cross or overlap each other. This embarrassing phenomenon is called quantile crossing and it has long been one of the challenging problems in the literature. We address the quantile-crossing problem using SV regression approach. With the common use of kernel trick, we derive a non-crossing conditional quantile estimator in the form of a constrained maximization of a piecewise quadratic function. We also propose its efficient Plait's SMO like implementation by exploiting the specific property of the problem.

Fuzzy modeling is known as one of the effective methods to identify unknown non-linear input-output relationships. In gathering information from constructed models or constructing models from known information, the model's understandability becomes essential. This paper defines new axes by fitting distributed data in input space and proposes a fuzzy modeling method considering data structure. This paper calls these axes, "fusion axes". The effectiveness of the proposed method is shown through some numerical experiments.

We consider regression problems under asymmetric or/and non-constant variance error. We see this problem in several fields such as insurance premium estimation, medical cost analysis, etc. Applying the method of Least Squares (LS) to this problem yields unstable solution because of outliers that appears on one side of regression surfaces. Conventional robust techniques to deal with outliers, which intend to discard or down-weight the outliers equally from both sides of regression surfaces, does not help for asymmetric error. In this paper, we propose an robust regression estimator (an estimator of the conditional mean) under asymmetric or/and non-constant variance error by simultaneously training conditional quantiles in multi-layer perceptron (MLP). This is considered as a kind of learning from hint or multitask learning approach, i.e., we train the conditional quantile estimator as hints or extra tasks to improve generalization properties of the conditional mean estimator. Numerical experiments and an application to medical cost estimation problem have shown that our proposal has robustness and good generalization properties.

The search performance of evolutionary algorithms (EAs) has been widely studied. Interactions between genes in a chromosome, called "epistasis", make the theoretical investigation difficult. The goal of this study is a mathematical analysis of the effects of bacterial mutation on a bacterial evolutionary algorithm (BEA). The NK-landscape problem is employed for the investigation of this analysis in this paper. The search ability of bacterial mutation is formulated and compared with those of conventional mutation operations. It is shown that the bacterial mutation surpasses conventional ones in search performance.

Fuzzy models describe nonlinear input-output relationships with linguistic fuzzy rules. A hierarchical fuzzy modeling is promising for identification of fuzzy models of target systems that have many input variables. In the identification, (1) determination of a hierarchical structure of submodels, (2) selection of input variables of each submodel, (3) division of input and output space, (4) tuning of membership functions, and (5) determination of fuzzy inference method are carried out. This article presents a hierarchical fuzzy modeling method with an uneven division method of input space of each submodel. For selecting input variables of submodels, the multiple objective genetic algorithm (MOGA) is utilized. MOGA finds multiple models with different input variables and different numbers of fuzzy rules as compromising solutions. A human designer can choose desirable ones from these candidates. The proposed method is applied to acquisition of fuzzy rules from cyclists' pedaling data. In spite of a small number of data, the obtained model was able to give detailed suggestions to each cyclist. © 2002 Wiley Periodicals, Inc.

Fuzzy modeling is a method to describe input-output relationships of unknown systems using fuzzy inference. Interpretability is one of the indispensable features of fuzzy models. This paper discusses the interpretability of fuzzy model with/without prior knowledge about the target system. Without prior knowledge, conciseness of fuzzy model helps humans to interpret its input-output relationships. In the case where a human has the knowledge in advance, an interpretable model could be the one that explicitly explains his/her knowledge. This paper defines the conciseness of fuzzy models, and formulates the conciseness measure. Experimental results show that the obtained concise model has the essential interpretable feature. The results also show that human's knowledge changes the most interpretable model from the most concise model.

A cerebellar model arithmetic computer (CMAC) is a neural network whose advantage is fast learning. It is, however, difficult to decide on the various parameters of a CMAC in advance. The pseudo-bacterial genetic algorithm (PBGA) is an evolutionary algorithm that is efficient in local searching. This paper proposes a "PBGA/CMAC" system that decides the positions of partitions, which are the main parameters of a CMAC, using the PBGA. The PBGA/CMAC hardware is implemented for acceleration, because PBGA/CMAC needs a large amount of computation time. An efficient learning method using pipelining is also presented. It is found that the accuracy with the proposed learning method is almost the same as that of the conventional CMAC's learning. The PBGA/CMAC hardware is 140 times faster than that of the equivalent PBGA software.

This paper is dealing with the fuzzy clustering method which combines the deterministic annealing approach and the fuzzy entropy. Regularizing fuzzy c-Means with the fuzzy entropy, we obtain the Fermi-Dirac type distribution function as the membership function. Then we examine the meanings of Fermi-Dirac function and the fuzzy entropy from the statistical mechanical point of view, and show that the mechanism of this fuzzy clustering is no other than the Fermi-Dirac statistics. Also we propose the algorithm which performs deterministic annealing of the membership function. Numerical experiment demonstrates that the proposed method can cluster a given set of data points properly.

Soft computing is a consortium of methodologies, which deal with various kinds of inaccuracies and uncertainties contained in real-world problems. In this paper soft computing is discussed from the viewpoint of a combination of characteristic features unique to fuzzy, neural, and evolutionary computing. This paper is intended to describe particularly the knowledge acquisition from inaccurate and uncertain data and to point out the areas where soft computing can become a breakthrough technology.

Fuzzy inference has a multigranular architecture consisting of symbols and continuous values, and this architecture has worked well to incorporate experts' know-how into fuzzy controls. The paper focuses on the important characteristic of fuzzy control, "symbolic expression" of a fuzzy control system. The paper first introduces a "chain of rules" to clarify the description of the behavior of the control systems. A coincidence of the symbolic and continuous behaviors is defined. With these definitions, the paper proposes a new condition, "relaxed nonseparate condition" and a new fuzzy inference method. The relaxed condition together with the new inference method guarantees the coincidence of symbolic and continuous behavior of the control system. The paper proposes a new design method of fuzzy controller that guarantees this coincidence. Simulations are done to show the feasibility of the relaxed nonseparate condition and proposed design method.

Fuzzy modeling is a method to describe nonlinear input-output relationships. Genetic algorithms (GAs) have been used with fuzzy modeling for identification of the structure of a fuzzy model and selection of input variables. Users often require fuzzy models that satisfy multiple evaluation criteria. Assignment of appropriate weights on these criteria is one of the key factors for good GA search. In order to give a guideline for assigning the degree of importance to each evaluation criterion for generating fuzzy models, we examined the characteristic of each evaluation criterion.

Stability analysis of fuzzy control systems is one of the main topics of fuzzy control. Previously, a stability analysis method on a symbolic level has been proposed. This method opened the door to the design of a stable fuzzy controller using symbols. The validity of the stability analysis in the symbolic system is not guaranteed in the continuous system. To guarantee this validity, a "relaxed non-separate condition" has been proposed. The problem is that this relaxed condition is only applicable to the systems that have symmetric membership functions. This paper introduces a new fuzzy inference method that enables us to use asymmetric membership functions under the relaxed condition. The proof of the related condition with the new fuzzy inference method is given in this paper.

Studies of the integration of patterns and symbols have attracted many researchers. Patterns are feature vectors in a feature space and symbols are labels of sets of patterns. The integration of patterns and symbols treats pattern-symbol pairs for information processing. Fuzzy inference is the integration of patterns and symbols by treating pattern sets as fuzzy sets and symbols as the labels of fuzzy sets. Fuzzy inference is very useful for extraction and incorporation of human knowledge. However, relationships between symbols and multi-dimensional pattern sets are hard to grasp. The paper proposes a fuzzy inference system with a nonlinear projector for dimensionality reduction to visualize relationships between the symbols and the multi-dimensional pattern sets. The proposed system is a good tool for interactive knowledge acquisition/incorporation.

This paper presents a new method for discovering the parameters of a fuzzy system; namely, the combination of input variables of the rules, the parameters of the membership functions of the variables, and a set of relevant rules from numerical data using the newly proposed bacterial evolutionary algorithm (BEA). Nawa et al. (1997) proposed the pseudobacterial genetic algorithm (PBGA) that incorporates a modified mutation operator called bacterial mutation, based on a biological phenomenon of microbial evolution. The BEA has the same features of the PBGA, but introduces a new operation, called gene transfer operation, equally inspired by a microbial evolution phenomenon. While the bacterial mutation performs local optimization within the limits of a single chromosome, the gene transfer operation allows the chromosomes to directly transfer information to the other counterparts in the population. The gene transfer is inspired by the phenomenon of transfer of strands of genes in a population of bacteria. By means of this mechanism, one bacterium can rapidly spread its genetic information to other cells. Numerical experiments were performed to show the effectiveness of the BEA. The obtained results show the benefits that can be obtained with this method.

This paper presents a study on the effect of transfer of genes for bacterial evolutionary algorithm. The bacterial operation, in the pseudo-bacterial genetic algorithm (PBGA) is efficient in improving local portions of chromosomes. In the experiments presented a gene transfer operation is introduced in the place of the crossover operation in the PBGA. The gene transfer operation allows the chromosomes to directly transfer information to the other counterparts in the population. The gene transfer is inspired by the phenomenon of transfer of strands of genes between bacteria that occurs in nature. By means of this mechanism. One bacterium can rapidly spread its genetic information to other cells. Numerical experiments were performed to show the effectiveness of the PBGA with the gene transfer operation. The obtained results show the benefits that can be obtained with the newly introduced operation.

Fuzzy modeling is a promising technique to describe input-output relationships of nonlinear system. This paper presents a new hierarchical fuzzy modeling method using genetic algorithm (GA). Uneven allocation of membership functions in the antecedent of each submodel in the hierarchical fuzzy model can be achieved with the proposed method. This paper introduces a simple coding method and a quick rule identification method for efficient search for a submodel using a fuzzy neural network (FNN). The obtained hierarchical fuzzy model are more concise than those identified with the conventional methods.

The architecture constructed with two types of processing, symbolic processing and parallel distributed processing, seems promising for intelligent systems. But the "symbol grounding problem", the discrepancy between the system and the symbols given by the top-down approach, arises. To solve this problem, the paper presents a framework in which the system self-organizes grounded symbols and uses them for symbolic processing. A navigation task of an autonomous robot is employed as an example. The state space for the robot is self-organized by the distributions of the input-output relationships through the interactions between the robot and the environment. Using the self-organized symbols, the robot can navigate by switching some neural controllers autonomously. Some simulations show the effectiveness of the proposed method.

Fuzzy inference is a method to describe nonlinear input-output relationships using fuzzy if-then rules. Continuous values of the inputs and outputs are converted into granules by fuzzy sets, and each granule is labeled with a symbol. Fuzzy inference has a multigranular architecture consisting of continuous values and symbols, and this architecture has worked well to incorporate experts' know-how into fuzzy controls. One of the important problems of fuzzy control is to guarantee stability of the fuzzy control system. The authors have applied Petri nets to the stability analysis of the fuzzy control system. A theory of asymptotic stability has been derived for the symbolic representation of the control system. The paper presents a new method to bridge between the stability analysis on the symbolic level and the actual behavior of the control system on the numerical level. The new method uses a generalized fuzzy Petri net model and its neural network representation. The paper introduces a guideline for designing a fuzzy control system which guarantees the validity of the stability analysis on the symbolic representation of the control system.

Presents an adaptive digital filter based on the multi-agent concept. The proposed filter consists of nonlinear small filter agents which cooperate with each other. The proposed filter is capable of adapting to a dynamically varying environment by changing the roles of the agents. At an instant, an agent is assigned to act as the main agent to minimize the error between the filter output and the teaching signal. The other agents, which assist the main agent, aim at minimizing the error of the main agent. Experimental results show good performance compared with conventional filters. The proposed filters are capable of adapting themselves to dynamically varying environments, while the conventional methods fail.

Presents a method for discovering the parameters of a fuzzy system, namely the combination of input variables of the rules, the parameters of the membership functions of the variables and a set of relevant rules, from numerical data using the newly proposed bacterial evolutionary algorithm (BEA). In early work, the authors proposed the pseudo-bacterial genetic algorithm (PBGA) that incorporates a modified mutation operator called bacterial mutation, based on a natural phenomenon of microbial evolution. The BEA has the same features of the PBGA, but introduces a new operator, called gene transfer operation, equally inspired by a microbial evolution phenomenon. While the bacterial mutation performs local optimization within the limits of a single chromosome, the gene transfer operation allows the chromosomes to directly transfer information to the other counterparts in the population. The gene transfer is inspired by the natural phenomenon of transfer of strands of genes between bacteria in a population. By means of this mechanism, one bacterium can rapidly spread its genetic information to other cells. Numerical experiments were performed to show the effectiveness of the BEA. The obtained results show the benefits that can be obtained with the newly proposed method.

This paper proposes a fuzzy abductive inference with degrees of manifestation. The fuzzy logic is applied to Peng and Reggia's abductive inference for handling the manifestation degrees. This method infers irredundant combinations of candidates with degrees of belief for the manifestations. A learning algorithm for updating the fuzzy causations and t-conorm parameter is also presented in this paper. Application of the new method to a diagnostic problem is shown and the effectiveness of the proposed method is demonstrated. © 1998 John Wiley & Sons, Inc.

The authors present a new network which consists of symbol layers and a pattern layer for acquiring concepts and inferring meanings of patterns and symbols. Nonlinear dynamics which can cause chaotic vibration in the internal state of the network is used for the recollection of various related patterns and symbols. The authors also propose a multi-module network with the multiple symbol-pattern networks for association and inference from vague patterns with multiple meanings. This paper presents an attention mechanism for this model. This new mechanism works to control the search area for related patterns and symbols. Simulation using face patterns consisting of eyebrow, eye and mouth patterns are done to show that the attention mechanism works well to recall related facial expressions successively.

Hierarchical fuzzy modeling using fuzzy neural networks (FNN) is one of the effective approaches for modeling of nonlinear systems. Decision of antecedent structures of fuzzy models of nonlinear systems is made possible by a combination of FNN and genetic algorithm (GA). The disadvantage of this fuzzy modeling method is that the learning of FNN is time consuming. This paper presents an efficient fuzzy modeling method using simple fuzzy inference. The results of fuzzy modeling are heavily dependent on evaluation criteria. This paper also studies effects of evaluation criteria for the decision of the antecedent structure. Numerical experiments are done.

The procedure for acquiring control rules to improve the performance of control systems has received considerable attention previously. This paper deals with a collision avoidance problem in which the controlled object is a ship with inertia which must avoid collision with a moving object. It has proven to be difficult to obtain collision avoidance rules, i.e., steering rules and speed control rules, which coincide with the operator's knowledge. This paper shows that rules of this type can be acquired directly from observational data using fuzzy neural networks (FNNs). This paper also shows that the FNN can obtain portions of the fuzzy rules for the inferences of the static and dynamic degrees of danger and the decision table based on the degrees of danger to avoid the moving obstacle.>

A fuzzy modeling method using fuzzy neural networks with the backpropagation algorithm is presented. The method can identify the fuzzy model of a nonlinear system automatically. The feasibility of the method is examined using simple numerical data.