The present study considers a fuzzy color system in which three input fuzzy sets are constructed on the tone triangle. For a given fuzzy input, this system outputs the center of gravity of the three weights that are associated with respective grades. Two fuzzy sets (not black and white) are applied to the tone triangle relationship. By evaluating the attributes of chromaticness, whiteness, and blackness, a target color can be easily obtained as the center of gravity of the resulting fuzzy set. The output of the system is a tone triangle, which includes a compound vector with three weights (scalars) in color space. The difference between fuzzy input and the resulting inference output is shown by the input-output characteristics (linear shape and right triangle shape) between the chromaticness, the whiteness, the blackness of the input and the chromaticness of the output. Shape effects of conical fuzzy set are investigated.

View full abstract

It is one of the fundamental works to read the Japanese historical documents in Japanese historical studies. Most of these documents are written with "kuzushi-ji" characters and reading the documents needs considerable labor even for experts. Because of this background, we have been developing the Japanese historical character recognition system for the reading support system for Japanese historical documents. This recognition system demonstrates mostly good recognition accuracy but it may fall in case of recognizing characters which are blurred and also have fractures. In this research, we apply "SOM template" which is used in a recognition of ornament letters to the historical character recognition and try to reduce decline of the recognition accuracy.

View full abstract

We propose the method of digitizing process that clearly has not been established for the analysis of EEG Data. With the aim to advance the analysis of the EEG data, we have examined the differential processing method of temporal waveform data obtained from the EEG.

View full abstract

The present metabolic syndrome judgment is as follows: It is judged as metabolic syndrome if checked values are exceeding standard values. It is identified by the non-metabolic digitally if checked values are not exceeding standard values. We established a non-ill area among the values, and the values are judged more finely. The method was applied to other items of the checkup judgment sheet, too. They are glucose metabolism, liver function, renal dysfunction，blood test, and inflammatory response. All these 6 items including the metabolic syndrome are included as the judgment items of Dr. Ningendock. In this paper, the further improvement of Dr. Ningendock is explained.

View full abstract

In the study of the computational neural networks on learning and skill acquisition of the motion functions, MOSAIC model using the internal model proposed by Kawato is useful. The internal model based on the control model of Allen-Tsukahara is able to express physical motion by associativity of the feedback control and the feedforward control. In this paper, we give a subject a task that he/she should push a number of the button to be equal to a computer screen in order to confirm a change function between the feedback control and the feedforward control of the internal model. If the repetition motion is given to the subject several times, the feedforward control comes up conspicuously, and the ratio of the feedforward control for the feedback control rises. We discuss here the movement, the response time, and the electroencephalographic signal in this repetition control of vision-motion task. We also discuss a MOSAIC model using the neural network from the results of vision-motion task.

View full abstract

In this research, social learning and individual learning are focused as learning methods for task execution. Our purpose is to analyze brain activity difference of each learning method. Subjects belongs individual learning group, social and individual learning group and non-learning group to compare difference by learning methods. In addition, these 3 groups were measured brain activity while performing the Tower of Hanoi in MRI. In this report, we explain fMRI analysis results and performance results on task execution.

View full abstract

The present longitudinal study aimed to investigate the relationship between acquisition of performance skills and flow as an index of intrinsic motivation. At the end of each piano practice, partic- ipants completed the questionnaire about the flow experience scale during piano practice. To determine the effects of training with practice, several tests to assess performance skills were carried out before and after twenty hours of practice. The specific state in which the flow was experienced or flow experience itself were suggested to be effective in acquisition of performance skills.

View full abstract

This paper shows the result of the application of P300 speller to actual patients of ALS (amyotrophic lateral sclerosis). P300 speller, which uses P300 as the feature quantity and enables users to select letters just by thinking, is expected to be developed as a communication tool for seriously paralyzed patients like those with ALS. The aim of this study is to investigate the practicality of the actual application of the Japanese P300 speller to ALS subjects. In inputting letters with P300 speller, it is needed to input letters correctly in a short time to reduce user’s burden. It is also needed to do machine learning before use (pre-training). In the experiments, we applied RB-ARQ (Reliability-Based Automatic Repeat reQuest) and incremental learning method which the authors have been proposed. It is shown that discrimination speed can be improved with a small loss of the accuracy by applying RB-ARQ, which can reduce the number of stimuli to ALS subjects. This paper also shows that the incremental learning can reduce the pre-training time keeping the accuracy.

View full abstract

The authors measured electroencephalograms (EEGs) from subjects on recalling ten types of images presented on CRT. Each presented image consisted of ten types of line drawings of body part, tetrapod, home appliance, and fruits. The canonical discriminant analysis was applied to these single trial EEGs. Four channels of EEGs at the right frontal and the temporal area were used in discrimination. They were Fp2, F4, C4 and F8 according to the international 10-20 system. Sampling EEGs were taken from 400ms to 900ms at 25ms intervals. Also, data were resampled -1ms and -2ms backward. The number of variates is twenty one by four; so the data were eighty four dimensional vectors and number of the data was three hundred and sixty. Results of the canonical discriminant analysis by use of so called jack knife statistical method were almost 80 % in case of ten types. These results were improved from our precedent research.

View full abstract

We have developed "Air Brain" system, an easily EEG measurement system with a smartphone. This system is consisted of general-purpose components and enables us to develop convenient system for EEG and human behavior telemetry. We are developing the BCI systems utilizing Air Brain system. The present study, as an example of these BCI systems, attempted to develop a new BCI system for controlling the camera via WiFi-Direct. By using the system Air Brain, we are able to develop useful and frequent BCI systems only by creating smartphone applications.

View full abstract

Recently, with the evolution of advanced information society, robot will play a significant role for supporting information service in the welfare industry. When the robot partner communicates with human, the robot perform the gesture based on the information in order to transfer it to human clearly and reliably. In this paper, we propose a gesture generation support system for the robot partner using Interactive Genetic Algorithm. Finally, the performance of gesture generation system is evaluated by using Laban feature vector and discuss the validity.

View full abstract

In order to elucidate higher functions of brain, it is critical to analyze changes electrical activity in a neuronal network quantitatively. We have developed the neuro-robot, which moves according to the response patterns of the neural network evoked by electrical inputs from outer world. In this study, we utilized Self-Organization-Map (SOM) to generate robot behavior without interception of teacher learning. According to IR sensor value of the robot, the neuronal network stimulated using a specific electrode, and a 64-degrees-feature vector of spatiotemporal pattern of neuronal electrical activity was generated. The values of these 64 nodes are inputted to two-dimensional output layer with 20×20 nodes of SOM. The SOM performed teacher learning only for the initial process in order to map different electrical patterns evoked by different inputs to distinct nodes. As a result, neuro-robot was succeeded in performing collision avoidance behavior.

View full abstract

Neuronal networks, composed by a lot of mutually connected neurons, contribute to highly functions of brain such as memory, learning and so on. Neuronal activity has reproducible spatio-temporal patterns and analyzing transition of the dynamical patterns is considered to critical for elucidation of information processing in brain. Modification of neuronal activity patterns after an electrical input stimulation by external world possibly linked to the transition of internal states in a neuronal networks. In this study, we attempted to classify spatial-temporal patterns of cultured rat hippocampal network on a multi electrode array (MEA) dish that has 64 planner electrodes on the bottom. We calculated compatibility degrees for 64 dimensional vectors from recorded neuronal electrical spikes by using fuzzy logic. Compatibility degrees were used for the criteria for a winner node in a 2-dimentional output layer of a Self-Organization Map, instead of Euclidean distances. We analyzed the transition of spatial patterns of spontaneous and evoked electrical activity in cultured neuronal networks, using this type of Fuzzy-SOM algorithm.

View full abstract

The information processing of the living neuronal network is embedded in neuronal network dynamics. Rat hippocampal dissociated culture on the multielectrodes-array (MEA) dish is useful as a simple model of a brain information processing system, because it possesses essential functions for information processing in the brain. In addition, analysis of neuronal network activity in the small network is also important for development of an interface between a human and a robot. In this research, we analyzed the property of pattern repertories in neuronal network activity, and estimated stability and periodicity of the network dynamic pattern. We measured neuronal network activity from 64 microelectrodes of MEA dish and counted the number of spikes in 5 ms - time windows and constructed 64 degrees feature vectors from the number of spikes in 64 sites. The number of clusters of the feature vectors in 30 min recording was approximately 1000 in spontaneous activity and 600 in evoked activity. However, the number of clusters with over 1 % reproducibility was approximately 15 in both of spontaneous and evoked activity. In addition, we estimated number of the feature vectors with same patterns (reproducible patterns). As a result of these analyses, we confirmed that the same pattern of electrical activity repeated in a long-term culture of living neuronal network.

View full abstract

In literature of network science, a "community" refers to a group of nodes that are densely connected within the group but are sparsely connected with nodes outside the group. Community structure is a hallmark of a variety of social, biological and engineering networks. Especially in neuroscience, there have been growing interests in analyzing community structure of brain networks, which is expected to reveal functional modules in information processing in the brain. Community structure of brain networks is characterized by the following properties: Each community has some overlaps with others; communities are hierarchically organized. Most of algorithms proposed up to now fail to detect community structure with these properties. Here we put forward a method for detecting overlapping and hierarchical community stricture of networks. Community detection from C. elegans neuronal and macaque cortical networks by this method is examined.

View full abstract

This study argues the mechanism of figure-ground organization using geometric information of objects' contour in the human visual system based on the psychological and physiological findings. Previous model proposed by the author can detect figural side of arbitrary closed contours, but cannot explain human hole-perception for a kind of contours. This study proposes the revised model which can mimic the hole-perception. The effectiveness of the model revision was confirmed by computer simulation.

View full abstract

The neocognitron is a multi-layered convolutional network that can be trained to recognize visual patterns robustly. This paper discusses training methods for the neocognitron. The add-if-silent rule is used for training intermediate layers of the hierarchical network. By the add-if-silent rule, a new cell is generated when all postsynaptic cells are silent. The generated cell learns the activity of the presynaptic cells in one-shot, and its input connections will never be modified afterward. At the highest stage of the hierarchical network, the method of interpolating-vector is used, not only for classifying patterns during recognition, but also for the training. In the highest stage, each feature-extracting cell is given a label indicating the class name of a training vector, and its preferred feature is represented by a reference vector in the multi-dimensional vector space. We assume plane segments spanned by every trios of reference vectors of the same label. During the recognition phase, the label of the plane segment nearest to the test vector shows the result of classification.

View full abstract

With the growth of robot technology, more educational-support robots, which support learning, are paid attention to. However, there is a problem in these robots. The problem is that a user loses his/her interest in the robot. To solve this problem, some studies which focus on a model of emotional expressions have been reported in Human-Agent-Interaction. A model of emotional expressions is defined as the agent expressing his/her emotion as with autonomous emotions. These models were shown to be beneficial to effective interaction between the agents and human. Thus, this paper examines how psychological effect of robot which expresses the emotion by using the model of emotional expressions can prompt learners in a collaborative learning.

View full abstract

In this paper, we propose the sympathy expression model of the bystander robot which adopted a concept of the moral emotion. The multiple researchers proposed various emotion generation models of the robot to date. However, most of these models are designed assuming one to one communication. Therefore these models can't return a natural reaction in group communication. Accordingly, we pay attention to the robot which is in a position of the bystander who is unrelated to the communication between participants. We propose a sympathy expression model that a bystander robot learns the emotional display of others and enables cooperative expressiveness. This model enables the appropriate expressiveness of a robot affecting communication in the situation of the bystander. Therefore we assume the interaction of three robots with the emotion generation model using the neural network. Further we inspect the movement of this model by using a psychology model. As a result, we confirmed the appropriate movement of this model.

View full abstract

Recently, the robot technology which can realize the smooth communication with human has been required. In this study, we obtain the time-series data of the physical movement including the human emotion by using Recurrent Neural Network (RNN). And, we developed a system which is able to judge his/her emotion considering a time series of flows of the physical movement. In addition, we report that we confirmed the advantage of obtained RNN by performing the simple experiment of emotion estimation.

View full abstract

In this paper, we propose an emotion generation model for robots that distinguishes man's personality. The robot performs different reaction by way of human communication with the robot. Most studies of the communication robot came to be conducted lively recently. In several studies, robots generate reactions to man by judging man's interaction to the robot. However, if a man who communicates with robots replaces another man and both of input to the robot are the same, the robot's reaction will become the same. To solve this issue, we proposed an emotion generation model for robots that performs different reaction by way of the robot distinguishes man's personality. In our proposed model, the robot has the degree of intimacy which judges man’s personality. When a man satisfies robot's desire ， degree of intimacy becomes high. On the other hand, when a man doesn't satisfy robot's desire, degree of intimacy becomes low. In simulation results, we confirmed that the robot shows friendly reaction to a man with the high degree of intimacy and shows unfriendly reaction to a man with the low degree of intimacy.

View full abstract

We have proposed the color combination system using the emotion color combination model based on the quantification theory type I and the image emotion model based on the quantification theory type II. Also, a uniform color combination system was constructed using the proposed method. In this method, one emotion was selected by the image emotion model. However, it was problem that only one of the emotions is selected, because it may be determined emotions intermediate depending on the selected image words. In this paper, we try to extend the fuzzy image emotion model, and a degree of membership to each emotion is calculated. Also, we try to construct color combination system that can display color combination candidates of emotion of multiple in accordance with the degree of membership.

View full abstract

This paper describes an extraction method of characteristic facial expressions for monitoring system of elderly people. The past few years, the number of elderly people living alone has a tendency to increase, therefore various systems for an elderly monitoring or safety confirmation has been realizing and utilizing. In the field of robots, there are studies of the monitoring system using robots at an elderly home. In the future, many systems are considered to advance into markets. This proposed method is to extract several characteristic facial expression images from video image, to estimate their rough psychological state rather than to convey safety information of elderly people as conventional. Specifically, facial expression images are extracted by the index of the difference of features between criteria's face image and face images detected from each frame.

View full abstract

This research is personifying a car using a lifelike agent, and aims at building a new relationship between a driver and a car. We develop a personified car agent who performs the suitable interaction for a driver by using operating information with the OBD system which acquires the information about the operation of a driver from a car in real time. In order to build the suitable interaction model to a driver, we conducted the experiment which verifies acquisition of the individuality information which appears in operation from the information on accelerator valve travel, speed, and engine number of rotations. We introduce the contents of analysis and discuss an agent design for the construction of new relationship between a car and a driver.

View full abstract

This paper describes a 3D plane detection method for an intelligent robot to detect unknown objects based on an information of 3D range sensor. Previously, various methods have been proposed to perceive unknown environments. However, previous unknown object detection methods have problems which are high computational costs and low-accuracy for small object detection. In order to solve these problems, we propose an online processable unknown object detection method based on 3D plane detection. The proposed method consists of a simple plane detection that applies particle swarm optimization (PSO) with region growing (RG) and an integrated object plane detection. The proposed simple plane detection is focused on to detect small planes and to reduce computational costs. To improve the accuracy of plane detection, we apply PSO and RG. Furthermore, an integrated object plane detection focuses on the stability of the detecting plane. Through experiments of object detection, we show that the computational costs are reduced drastically for real time operations. Furthermore, we show the improvement of small plane detection accuracy.

View full abstract

In this study, we aimed to proposal of the decision making support system for mobile phone replacement demand. The system shows simple questions about selection criteria of product to users. Thus, the system could recommend the product that matches the user's preference. First, we adopted the Fuzzy AHP that can reflect people's subjectivity. Because, we think that it can be flexible decision making support. And, we set the evaluation value of alternative (mobile phone) by the paired comparison method. We think that our system can support the answer of decision making problem in the mobile phone replacement demand. So, users would have confidence to the answer. And, they would be able to do better selection of mobile phone. In this paper, we show the experiment and results using the Fuzzy AHP. And, we propose and verify the decision making support system.

View full abstract

Japanese auto companies maintain their competitive advantages in the world. Additionally it is deduced that dominant design in the industry changes from conventional petrol vehicles to new generation vehicles such as Hybrid Vehicle (HV). Our purpose is comparisons of the three companies focusing on change of auto industry caused by HV unlike other conventional researches comparing Japanese auto companies at the corporate level. Specifically, this paper visualizes and compares their technology innovation strategies in the HV industry by using their patent documents as their ample source of technical and commercial knowledge.

View full abstract

This paper introduces two strolling support systems that utilize ambiguous information. The first system alerts users to the existence of nearby attractive streets and areas by vibration and visualized characteriza- tion of streets while the second system notifies users of high-frequency photography areas by vibrating the smartphone. The systems only display the distribution of the areas on the screen; they provide no detailed information such as maps and routes. In user studies, 16 participants examined their use while walking around the city. The user studies suggest that the ambiguous information provided by the systems attracts users' attention to surrounding environments and increases the amount of photography and conversation during their walk.

View full abstract

In this paper, we propose a system to create harmonic designs that reflect preferences of multiple users with votes using digital signage. When creating harmonic designs, it is important to collect a lot of opinions of users and to adopt them for designs. However, as the number of users increases, the collection of opinions is more difficult. Then, we propose an Interactive Evolutionary Computation (IEC) system using digital signage. The proposed system presents a lot of designs for multiple users by digital signage, and each user evaluates designs by voting. In other words, the proposed system obtains preferences of multiple users by voting and uses them to evaluate for designs. Therefore, the proposed system is able to create harmonic designs that multiple users are pleased. In our previous research, we have proposed Paired Comparison Voting (PCV) method that obtains preferences of multiple users with votes and confirmed the effectiveness by simulation. The proposed system uses PCV method to create designs. In this paper, we construct a system using PCV method and digital signage and demonstrate the effectiveness for real users. The experimental results showed that the designs generated by the proposed system converged visually and genetically. As a result, we were able to verify that the proposed system can create designs that reflect preferences of multiple users.

View full abstract

This paper proposes a melody impression model that uses scale and key for an automatic composition system that reflects sensibilities Kansei selected by users. There are many modes of scale each gives a different impression to listeners. Modes and also impression of a given scale depend on its notes and their sequence. Also, the membership of a given melody to each scales are fuzzy. The authors model a difference between the impressions by Kansei evaluation of 25 randomly generated melodies. Then we analyzed the experimental data of Kansei evaluation by quantification theory I. Finally, we construct a melody impression model based on the results of analysis.

View full abstract

This paper proposes a support system for logotype design based on impression of a target logotype. The presented system has the impression estimate part, the logotype generation part and the logotype modification part. Especially, this paper describes the roles of the impression estimate part and the logotype generation part, and the procedure of generating an initial logotype design. Impression of a target logotype, which is expressed using an adjective, and a character string are inputted to the system. In the impression estimate part, the inputted adjective is analyzed by the concept of co-occurrence of adjectives and its impression value is obtained. The logotype generation part has an impression space that expresses a factor structure of the impression of a logotype, which is obtained by Factor Analysis of data in the pre-experiment. This part generates logotype candidate that has the lowest non- similarity measure of the impression value among the impression values of logotype samples in the impression space comparing with the impression value of the inputted adjective. The results of the subject experiment show that the initial logotype design generated by the presented method reflects impression expressed by an adjective well.

View full abstract

This research introduces several touch panel ad-hoc games to aid in therapy. The motivation is important point of patient to actively participate in the rehabilitation process. In this paper, the rehabilitation system is developed based on cognitive exercise therapy. This system use a prototype touch screen device. The usability evaluation test is employed to assess system's effectiveness. Additionally, two different touch screens were used, tested and evaluated.

View full abstract

In this paper, we verify the performance of Kansei model in Healthy Eating Habits Support System (HEHSS). A meal is very important for a human life. However, considering daily menu is a very labor-intensive task for people who cook every day. To solve this issue, we have developed the HEHSS considering a user's taste and health. In the HEHSS, NPD Kansei model expresses the user's taste. NPD Kansei model expresses the user's P (Positive: like) taste, N (Negative: dislike) taste, and D (Don't care: neutral) taste in each items that make up recipes, and evaluates recipes on behalf of the user. In order to consider the user's taste, NPD Kansei model must be able to express the user's taste appropriately. However, in our previous study, we have not confirmed whether NPD Kansei model is effective as a method of expressing the user's taste. Therefore, we verify the effectiveness of NPD Kansei model by performing evaluation experiments using some real users. The experimental results show NPD Kansei model is effective as a method of expressing the user's taste.

View full abstract

Macrophages are an essential part of our immune system and one of the well-known cells in biology. However, despite its importance its usual behaviour and moving rules are mostly unknown due to the lack of observatory data for the macrophages. An experiment was conducted in an attempt to obtain new data by observing the macrophages in a living rat and produced a couple of movies. Our objective is to exploit these results to develop a simulator that is able to reproduce the moves of a macrophage population. After the movie analysis, we identified a few characteristics to implement into a first simple simulator. Although its performance is very limited, it represents a first step for further research of understanding macrophages.

View full abstract

Mini4WD-AI is for the purpose of machine control with not mechanical control, but electrical control. In order to realize Mini4WD-AI, lightweight and small computer system with various sensors and motor drivers are needed. Several kind of technology is desired to realize it. We propose the hardware design method and show the simple motor controller.

View full abstract

The purpose of this study is to demonstrate the applicability of Cellular Automaton Particle Swarm Optimization (CAPSO) for constrained problems. CAPSO is a method that can efficiently search solutions with an autonomic action of each particle adapted to the applied problem. In the optimization for practical problems, a solution search requires the applicability to a design space including constraint conditions, various evaluation criteria and uncertainties. In past researches, search methods for certain problem such as the multi-objective optimization have been proposed. However, the improvement of those methods for application problems tends to become difficult due to the complication of their algorithm. On the other hand, CAPSO can treat the algorithm of search solution and the description of particle’s action in response to a problem separately. By this feature, CAPSO has the extensibility for practical problems. In this paper, an attempt is made to propose an action rule of particle suitable for constrained optimization problems in consideration of the extensibility of method. Numerical examples in the function optimization are presented to demonstrate the applicability of proposed method.

View full abstract

Evolutionary multiobjective optimization (EMO) has attracted much attention over the last decade for solving multiobjective optimization problems. To examine the behavior of EMO algorithms, many test problems have been proposed so far. Each problem has different characteristics. One important feature is the relation between the solution space and the Pareto front size. For example, the distribution of randomly-generated initial solutions is much larger than the size of the Pareto front. In such a case, the population should shrink and converge towards the Pareto front. In this paper, we examine the characteristics of well-known multiobjective problems (e.g., knapsack problems and DTLZ problems) with two to ten objectives.

View full abstract

In this paper, we proposed a novel Multi Agent System (MAS) according to SIR model for observing the process of epidemic. SIR model is one of the Mathmatical model, whose functions can obtain an approximate solution about infection rates. However, it has difficulty to represent the emergent phenomenon, therefore we used a MAS to solve this problem and created the simulator considering the agent's commuting route and his house. We examined a transition of the percentage of infected agents on several city models with this simulator.

View full abstract

We investigate the combinatorial effect of evolutionary multi-objective optimization (EMO) with interactive evolutionary computation (IEC). The purposes and combination ways of several presented EMO+IEC researches are different. We evaluated seven combination ways of four EMO objectives given by fitness functions and one IEC objective given by a pseudo-IEC user outputting stable evaluation regardless repeated experiments in our previous experiments. In this paper, we extend experimental conditions to 39 and evaluate them: 3 pseudo-users * 13 combination ways of 4+1 objectives. We also consider features of this system.

View full abstract

It is the important problem to prevent premature convergence in study of optimization algorithm. Especially, the adjustment of the threshold to select the useful solution is needed in the model using the Multi Agent System. We are developing a novel optimization algorithm inspired of the foraging behavior of the honey bee, which has the function to control the threshold. This is, in the algorithm, each agent has personal threshold to evaluate the usefulness of the solution, divides into exploration and exploitation of the solution and prevent premature convergence by adjusting the threshold autonomously. However, the study of the effects of time and space that are the most important factors in honey bee foraging behavior has not studied enough. So, as a preliminary stage of the development to the algorithm, we produced the model of honey bee foraging behavior.

View full abstract

Although the FCM clustering algorithms based on entropy or K-L information regularization approaches are often identified with statistical clustering model based on Gaussian mixture models, they have an advantage of achieving hierarchical clustering supported by deterministic annealing scheme. In this research, the applicability of deterministic annealing scheme is discussed in the fuzzy co-clustering context, in which a multi-nomial mixture models was enhanced to an FCM-type co-clustering model with K-L information regularization.

View full abstract

In order to achieve personalized recommendation by utilizing vertically distributed log data, individual records of the data sets must be summarized considering preservation of personal privacy. In this research, a privacy preserving fuzzy co-clustering algorithm is applied to real-world cooccurrence data and the applicability of the co-cluster structures in collaborative filtering is studied.

View full abstract

In k-Means-type clustering, clustering algorithms are often implemented with various parameter settings and the most plausible solution is selected supported by several validity indices. Xie-Beni index is a popular validity index in FCM clustering, which measures the plausibility level of fuzzy partitions con- sidering cluster compactness/separateness, and was modified for fuzzy co-clustering tasks. In this research, the applicability of the Xie-Beni-type co-cluster validity index is studied with several fuzzy co-clustering models such as cluster aggregation models (FCCM and Fuzzy CoDoK) and pseudo-cluster-center models (SCAD).

View full abstract

This paper proposes a new type of semi-supervised fuzzy c -means clustering with assignment prototype term. The assignment prototype term is based on the Windham's assignment prototype algorithm which handles pairwise constraints between objects in the proposed method. First, an optimization problem of the proposed method is formulated. Next, a new clustering algorithm is constructed based on the above discussions. Moreover, the effectiveness of the proposed method is shown through numerical experiments.

View full abstract

In this report, some fuzzy clustering methods are proposed based on Tsallis entropy-regularization. According to that conventional entropy-regularized fuzzy c-means (FCM) is obtained by regularzing Kmeans objective function with Shannon's entropy, Tsallis entropy-regularized FCM is proposed by regularizing Kmeans objective function with Tsallis entropy. The proposed method is different from Menard's method where Tsallis entropy is used not with Kmeans but with the standard FCM. Additionally, a variable controlling cluster size is introduced, and a maximizing model is proposed both for the proposed method and Menard's method.

View full abstract

Recently, kernel functions are incorporated into many clustering methods. There are, for example, Gaussian kernel, polynomial kernel, perceptron kernel and sigmoid kernel as representative kernels. There are, however, many data where we do not have adequate results of clustering by these kernels. It is necessary to research a new kernel that replaces existing kernels. In this paper, we study two kernels derived from fuzzy c-means and compare these kernels with conventional ones. We investigate classification performance and calculation time of the proposed kernels using illustrative examples.

View full abstract

Visualization methods are important tools to intuitively understand the relationships between each element in complex data sets. Until now, many visualization methods have been proposed and some of them are used in practice. However, the many existing methods can handle only symmetric data sets in which the relationship between each datum are symmetric and can not be applied to asymmetric data sets as is. Considering that there are many asymmetric data sets, the development of the visualization methods to handle the asymmetric data sets is a significant problem. This paper focuses on existing graph visualization methods based on a spring model and give a proposal to applying the model into asymmetric data set.

View full abstract

Methods of qualitative analysis such as qualitative classification have gained importance as an essential complement of existing quantitative analysis in numerous fields, such as behavior finance, econometrics, and business management. Only a few models have been developed to deal with qualitative inputs (attributes), which appear in the form of T2F data. Additionally, classification models are unsuitable if an output point is not fully assigned to a single class. In this paper, we formulate a comprehensive qualitative classification model based on fuzzy support vector machine (FSVM in brief) combined with Type-2 fuzzy expected regression (FER in brief) to deal with T2F inputs. This classifier(FER-FSVM in brief) makes it possible to achieve discrimination of output while characterizing membership for each class in terms of multi-dimensional qualitative inputs (attributes). Moreover, FER-FSVM can self-learn the data structure and shifted between FER or FSVM for classification automatically. It will largely shorten the computing time especially for large datasets by using linear structure of FER classifier to limit the size of non-linear classification region.

View full abstract

In this paper, we propose an approach to extract signboard regions from scene images. We first carry out toggle mapping which is able to detect a region whose intensities are homogeneous. We therefore detect an intensity homogeneous area by using toggle mapping. A support vector machine is applied to classify a homogenous area into signboard or non-signboard classes. Finally, we conduct computer experiments to show the effectiveness for the proposed method.

View full abstract

We are developing a head mounted display (HMD) which aids a low vision individual in his visual recognition (we may call it an "eyesight aid" taking a visual equivalent for "hearing aid"). We assume our targeting eyesight aid to be equipped with the following helping functions. 1. Visual recognition: it helps the low vision individual find out the object in his view which he wants to recognize or the one which he is better to recognize (for example, a pictogram of a exit). 2. Proper viewing: if he is of narrowed view or of lack of part of view, it navigates his view (gaze) point to the direction in which he could catch the object in the clear view region. 3. Easy seeing: it enhances the image on the display, for example, it exchanges black and white colors, adjusts contrast. In this paper we describe the development status of recognition software and low vision individual's HMD and HMD viewing area measuring method of low vision individual.

View full abstract