Robot technologies developed for industrial robot have recently moved its focus to service robot. Service robots are used to fulfill the purposes of improved comfort lives of humans with safe supports. In this lecture, I will explain a method of implementation of internal communication network of mobile robots based on the Optic- EtherCAT, industrial Ethernet.

抄録全体を表示

This paper provides a summary of the contents of the invited lecture at the conference and the most important references where readers can obtain more information on the topic.

抄録全体を表示

Kansei data are multi-dimensional data. It is difficult for an analyzer to interpret data whose dimensionality is higher than three because his/her vision is used only to one -- three dimensions. Visualization by reducing the dimensionality of Kansei data to less than or equal to three dimensions could help the analyzer to understand the data. For an effective visualization, definition of distances between data is important. For the definition, axes that form a space of Kansei data should be determined first. The choice could be questions or objects. A questionnaire uses several objects and many questions. Respondents are asked to answer each question one by one by marking on a rating scale. Questions are usually used as axes for multivariate analysis. Objects are another choice for the axes. By changing our viewpoint from different axes, new relationships between questions/objects could be found.

抄録全体を表示

Collaborative filtering is a technique for achieving personalized recommendations in a network community, and a promising collaborative filtering approach is based on user or item clustering. In the previous study, we proposed a sequential co-clustering model for extracting user-item groups, where users and items are mutually connected, and showed that the user-item connections support high recommendation ability. However, the computational cost of extracting user-item co-clusters is high because eigenvalue problems of a (user + item) square matrix must be solved. In the present study, we consider the reduction of computational cost by taking into account the characteristics of the input data matrix, the off-diagonal block elements of which are all zero.

抄録全体を表示

This paper investigates a decision problem of evaluating alternatives under multi-criteria. For several terms, the individual evaluations of each alternative under each criterion are recorded repeatedly. Then, as a final decision the collective evaluation over all terms and all criteria is obtained. In order to deal each term and criterion independently, the aggregations of individual evaluations are done by the approximation, differently from by the summation. The first step is to aggregate evaluations over all terms. From the possibility view, the aggregation is obtained as the upper approximation of individual evaluations so that the varieties depending on terms are reflected in an interval. The second step is to aggregate the obtained interval evaluations under over all criteria. From the core evaluations view, the aggregations are based on the lower approximation. Depending on the viewpoints of criteria and alternatives, two aggregation models are proposed.

抄録全体を表示

In this paper, we sophisticated our agent model and simulated multi-agent society. Agent is designed to compress its data by the eigenvector of its data. Additionally, in order to examine the relation between images of groups and communication network, agents' societies are investigated with several parameters. Thus, this method is also designed for human-like decisions in the learning process. The simulation results indicate our real societies, such as classes, companies and so on.

抄録全体を表示

DEA is a nonparametric method for finding the relative efficiency of DMUs (Decision Making Units). However, the most of these studies is based on the situation that objective DMU has competed with other DMUs as a group of relative comparison. There is a little study from the analyst aspect that manages the all DMUs which are in the state of cooperation. This paper proposed the DEA based method in order to control the activities when all DMU are in the state of cooperation. Proposed method has constructed following three steps: (1) employee's temporary assignment based on DEA evaluation and analyses of inconsistent characteristics, (2) setting of dissatisfaction degrees for each employees based on inconsistent characteristics, (3) employee's assignment based on each dissatisfaction degrees. Numerical study shows that the proposed method can control the activities of DMUs according to analyst's intention.

抄録全体を表示

A new approach to DEA-based decision making is proposed in this paper, in which 2-D visual feature maps are constructed for intuitively assessing efficiencies of DMUs (decision making units) by using results of data envelopment analysis (DEA). In order to summarize intrinsic information of DEA results efficiently, several variants of principal component analysis (PCA) such as fuzzy PCA or PCA with variable selection are adopted. With the goal of clear identification of target efficient DMUs, the feature map is constructed by emphasizing the mutual relation among efficient DMUs where relatively larger fuzzy memberships are assigned to them in fuzzy PCA. Variable selection mechanism is also applied in order to avoid the illegal influences of singular DMUs. The advantages of the proposed approach are demonstrated in a numerical experiment with a data set on Japanese prefectural activities.

抄録全体を表示

"Shodo", which is Japanese calligraphy, is a form of artistic writing used for writing the Japanese and Chinese language. This study proposes and develops a method of designing Japanese calligraphic fonts which contain artistic representations similar to actual Shodo art. We especially focus on kasure (a Japanese term referring to the breakup of brushstrokes), which is one of the most characteristic visual effects of Shodo, adopt an approach using example-based texture-synthesis in the field of non-photorealistic rendering to achieve high quality artistic fonts.

抄録全体を表示

Companies often carry out questionnaire(s) in order to design marketing strategies or to grasp the market trends. Recently, Web questionnaire survey becomes popular with the spread of the Internet in order for companies to reduce cost and to get a lot of questionnaire data. However, in the Web questionnaire survey, some respondents do not answer the questions seriously, because it is not done face-to-face and is done just for giveaway basically. The authors have tried to quantify the conflict of response based on the contradiction represented by the pair(s) of questions with opposite meaning. One of the most important matters in the quantification of the conflict of response is the selection of the pair(s) of opposing questions, which are actually regarded as opposite meanings by respondents. This paper defines the distance between questions which shows the similarity of questions for respondents and proposes a method to derive the pair(s) of opposing questions based on the distance between questions. This paper applies the proposed method to an actual questionnaire data and shows more appropriate opposing questions for respondents can be extracted than the prepared pairs of questions.

抄録全体を表示

Many 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 data for getting outlines or classifying them. The authors have proposed the 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 helpful enough for the analyzer to grasp the outlines of the texts and extract opinions from them. This paper presents a new visualization method for the HK Graph incorporating a grouping 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.

抄録全体を表示

Various colors are used for things that we use in daily life. A color combination is important for design and has a great influence on image or human kansei. When color combinations of signboards are handled, it is necessary to consider not only the color combination of signboard but also the color combination blending in with the landscape. However, it is difficult to evaluate blending in with the landscape. Also, generation of color combination considering it is difficult. Therefore, in this paper, the interactive genetic algorithm is used to the color combination support system of signboards. We try to construct the signboard color combination system with human subjective evaluation.

抄録全体を表示

Recently, the need of users to products is variable to their preference, and current users prefer to the personal customization of products. A human-friendly and easy design support system is required for non-expert users. We proposed interactive design support systems based on evolutionary computation and fuzzy evaluation, and applied the proposed method to the appearance design of robot partners. However, it is very difficult to obtain human evaluation. Therefore, we propose a method of visualizing functions and capabilities through computer simulations. Furthermore, we use iPhone simulator to evaluate the human impression based on direct interaction with the designed robot partner. Finally, we discuss the effectiveness of the proposed system through several simulation and experimental results.

抄録全体を表示

This paper proposes a document clustering algorithm based on formal concept analysis. This novel clustering algorithm uses mathematically order information among data. Conventional clustering algorithms deal with numeric data so that every non-numeric data are transformed to numeric form. Proposed algorithm generates word vectors from document data and define a context table. Then it calculates a concept lattice based on the word inclusion and exclusion relation among data. The experiments are conducted using 100 and 200 documents derived from Reuters-21578 text database. The result shows that our algorithm can classified the documents and also can assess the significance of each cluster.

抄録全体を表示

In this paper, two semi-supervised clustering methods are proposed, which are based on entropy regularized fuzzy c-means algorithm. First, two fuzzy c-means algorithms are introduced. The one is the standard one and the other is the entropy regularized one. Second, two semi-supervised standard fuzzy c-means algorithms are introduced, which are derived from adding loss function of memberships to the original optimization problem. Third, two new optimization problems are proposed, in which one is derived from adding new loss function of memberships to the original optimization problem and the other is derived from adding the loss function used in the latter semi-supervised standard fuzzy c-means algorithm.Last, two iterative algorithms are proposed by solving the optimization problems.

抄録全体を表示

Classification algorithm for discharge summary stored in hospital is proposed for advance the quality of medical care based on structured information. It provides similarity of morpheme word in discharge summary for each DPC (Diagnosis Procedure Combination) code, which the similarity is calculated based on SOAP (subjective, objective, assessment, and plan) information stored in electronic health record. Experimental classification results for 550 discharge summaries of 55 DPC codes taken in Kochi Medical School hospital show that accuracy of classification is 85% and that the same words are used in discharge summary of the same DPC codes. The proposed algorithm can be used for intercomparison study of words of discharge summary in Japan.

抄録全体を表示

Transformation methods from tree to lattice, in order to find quickly the spread out categories, and from lattice to tree, in order to restructure tree, are proposed. Translation from tree structure to lattice structure enables to aggregate categories which are spread out into different categories. Also, users can restructure another tree by translating from lattice to tree. Experimental comparison using data from Yahoo! Directory shows the effectiveness of the proposed methods and it is confirmed that the computation of the proposed method can be done in real time.

抄録全体を表示

In order to solve the state explosion problem of reinforcement learning in soccer game simulation, a state-space reduction method using fuzzy logic is proposed. We apply the proposed method that state-space will be reduced from several millions to several hundreds to soccer agent model of Robocup 2D soccer simulation system. Through the experimental comparisons between the proposed method and the conventional soccor agent without learning in 20 games, it is confirmed that the proposed method improves the performance of taking goal of soccer agent.

抄録全体を表示

In this paper; implantation of fuzzy logic controller for parallel hybrid electric vehicles (PHEV) is presented. In PHEV the required torque is generated by a combination of internal-combustion engine (ICE) and an electric motor. The controller simulated using the SIMULINK / MATLAB package. The controller is designed based on the desired speed for driving and the state of speed error. In the other hand, performance of PHEV and ICE under different road cycle is given. The hardware setup is done for electric propulsion system; the system contains the induction motor, the three phase IGBT inverter with control circuit using microcontroller. The closed loop control system used a DC permanent generator whose output voltage is related to motor speed. Comparison between simulation and experimental results show accurate matching.

抄録全体を表示

Ultra wideband UWB radars in the 24 GHz domain are currently under development for many automotive applications. Automotive UWB radar is attractive for application in short range automotive radar systems using 22 to 29GHz in order to improve road safety and realize intelligent transportation. By employing a direct conversion receiver, automotive UWB radar is able to meet size and cost reduction requirements. Direct conversion architecture poses a dc-offset problem, which corrupts velocity ambiguity in automotive UWB radar. In this paper, we proposed a mean-padding method to reduce zero-padding FFT error due to dc-offset in automotive UWB radar using direct conversion receiver. Zero-padding degrades vehicular velocity measurement probability. The results of computer simulation and experimentation using the designed RF module of the automotive UWB radar show that the presented method achieves a better velocity measurement probability.

抄録全体を表示

In this paper, we discuss a routing problem of a single grasp-and-delivery robot with unit capacity, which is working at an assembly line of printed circuit boards. The graspand-delivery robot arranges n identical pins from their current configuration to the next required configuration by transferring the pins one by one. We refer to such a transition from a configuration to the next required one as a cycle. The n pins support a printed circuit board from underneath in order to prevent it from overbending, while an automated manipulator embeds a number of electronic parts on to the printed circuit board from above. Each printed circuit board has its own pattern of circuit, and the required configurations for printed circuit boards to be processed are different from each another. Given an initial configuration of the n pins and a sequence of m required configurations for the printed circuit boards, the routing problem asks to find a transfer route of the grasp-and-delivery robot which minimizes the route length over all the m cycles. In this paper, by applying a weighted matroid intersection algorithm, we show that the cyclic routing problem is 2-approximable in polynomial time.

抄録全体を表示

As intelligent vehicles have been continuously developed through technology convergence, and their level of safety or convenience has been considerably improved, the ratio of electronic sub-assembly has increased, and the number of ECU has also increased. Moreover, the importance of safety has increased, and higher reliability and availability are needed in the system. This paper intends to suggest fault-tolerant dual system for enhancing availability and reliability of vehicle safety system by using redundancy system, and to develop application for operating and managing the system.. This system using hot-standby sparing technique similar to FT-SPARC structure is composed of master and slave processor modules, and is distinctive in that only one module plays a role of master according to the operation state of system, that is, normal or abnormal operation state.

抄録全体を表示

Hybrid electric vehicles (HEVs) are built using advanced technology to reduce exhaust emissions and increase fuel economy, which are the main objectives of the automotive industry today. Motors/controllers are the most fundamental components, especially in a series configuration, to produce and control the drive force for the hybrid vehicles. This paper presents a method of selecting the optimal drive motor/controller for a series hybrid vehicles, based on the requirements. The paper provides an overview of electric drive motors/controllers and Analytic Hierarchy Process. ADVISOR is utilised to simulate the hybrid vehicle performance, employing different categories of AC motors/controllers in the powertrain. The Analytic Hierarchy Process is then implemented with consideration given to several performance criteria for the comparable motors/controllers of the series hybrid vehicle. Results show a typical permanent magnet motor/controller has a better performance compared to induction motors/controllers.

抄録全体を表示

This paper has presented an acquiring method for agents' actions using Ant Colony Optimization (ACO) in multi-agent system. ACO is one of powerful meta-heuristics algorithms and researchers have reported the effectiveness of applications with the algorithm. We have developed the experimental system with soccer agents using ACO in RoboCup Soccer Simulation League and have reported the results of some experiments with our system.

抄録全体を表示

In this paper, we first discuss the notion that artificial market systems should meet the requirements of fidelity, transparency, reproducibility, and traceability. Next, we introduce history of development of the artificial market system named U-Mart system that meet the requirements well, which have been developed by the U-Mart project. Next, we propose a novel U-Mart system called "U-Mart system version 3.0" to solve problems of conventional U-Mart systems. The computational experiments are carried out taking a market simulation where the futures price affects the spot price to show the effectiveness of U-Mart system version 3.0.

抄録全体を表示

Q-Learning is a popular non-supervised reinforcement learning (RL) technique that learns an optimal actionvalue function characteristic of a learning problem. Due to the complexity of some problems, the number of training episodes to reach the convergence of the learning problem can increase drastically. In order to fasten the learning speed of an agent on a particular problem, researchers have been exploring interactive reinforcement learning (IRL), i.e. a way to interact with an agent so that it does not learn to solve a problem only by itself. This paper proposes an interactive reinforcement learning to try to fasten the learning speed of an agent. Especially, the combination of an agent asking for advice and getting advice from supervisors was explored. A simple way to experiment this combination is an agent evolving on a maze (a gridworld problem) trying to find its path to a fixed goal point. Experiments shows how an interactive learning agent solve the problem compared to a classical learning agent.

抄録全体を表示

Genetic Algorithms (GAs) have been widely applied to Multi-objective Optimization Problems (MOPs), called MOGA. A set of Pareto solutions in MOPs having plural fitness functions are searched, then GA is applied in a multipoint search. MOGA performance decreases with the increasing number of objective functions because solution space spreads exponentially. An effective MOGA search is an important issue in Manyobjective Optimization Problems (MaOPs), which are MOPs with four or five objective functions or more. One effective approach is aggregation of objective functions and reducing their number, but appropriate aggregation and the number of objective functions to be aggregated has not been studied sufficiently. Our purpose here is to determine the effects of aggregation of objective functions quantitatively. This paper studies the effects of aggregation with the number of aggregated objective functions based on the evaluation criteria proposed in this paper when MOGA is applied to a Nurse Scheduling Problem (NSP).

抄録全体を表示

This paper aims to accelerate learning processes of actor-critic method, which is one of major reinforcement learning algorithms, by a transfer learning. Reinforcement learning allows agents to work out the target tasks, autonomously. Transfer learning is one of effective methods to accelerate learning processes of machine learning algorithms. It accelerates learning processes by using prior knowledge from a policy for a source task. Two basic issues for the transfer learning are method to select an effective source policy and method to reuse without negative transfer. In this paper, we mainly discuss the latter. We propose the reuse method based on our proposed selection method. In actor-critic, a policy is constructed by two parameter sets: action preferences and state values. To avoid negative transfer, agents reuse only reliable action preferences and state values that imply preferred actions. We perform simple experiments to show the effectiveness of the proposed method.

抄録全体を表示

Recently, bag-of-visual-words has been studied as an image retrieval approach that uses the defining features of images. However, k-means clustering, generally used in bag-of-visual-words, has a drawback in that its results are affected by setting initial points and their number. Additionally, the more the number of keypoints increases, the more expensive processing becomes. We solve these problems of bag-of-visual-words by using a quantizing method that we have developed. In addition, we have developed a theme comprehending system that uses ontology.

抄録全体を表示

This paper shows the performance of glaucoma diagnosis of fundus images using the Subspace Classifier method. Feature extraction is based on three image color channels (R, G and B), and the optimal subspace dimensionality for the analysis of the fundus has been selected. A series of the analyses has been conducted, in which the classification accuracy for glaucoma in the fundus image has been compared between the proposed method and two other methods, namely the Learning Vector Quantization and Multi-Layer Perceptrons. In the experiments a maximum recognition rate of 75.3% was obtained by using the Subspace Classifier method, which was the best-performing one among the three methods compared.

抄録全体を表示

We present a robust semi-supervised method using the extended mode filter for learning with partially-labeled training data including label errors. The mode filter was originally developed for smoothing images contaminated with impulsive noises and usually needs iterative solution methods. In this paper, we propose a direct solution method with full search of solution spaces. This direct method outperforms the iterative algorithm in classification rates and computational speeds. Additional iterations of the mode filter raise up the classification rates. We extend the mode filter by introducing weights based on the isolation degree of data, and show the effectiveness of this extension by UCI benchmark data and UMIST Face Database

抄録全体を表示

This paper proposes a method for investi- gating users' impressions of products using a Fuzzy C4.5 decision tree. A survey was conducted through a questionnaire that asked subjects about their intuitive impressions of the appearance of running shoes. The answers to the questions were very subjective and indefinite. This subjectivity and vagueness could cause difficulty in constructing a reliable decision tree. In this study, we propose a method for obtaining a reliable decision tree and determining rules affecting user impressions. A decision tree was then built using questionnaire results to help uncover major factors affecting the product' s attractiveness. Moreover, we developed an application for the visualization of analytical results.

抄録全体を表示

An algorithm based on facet classification is proposed to reorder suggestions of resource annotations. It finds the most common pattern of facets used over data in order to reorder according to the suggested annotation class and its popularity. The algorithm is applied to a real, large dataset of a well-known online repository, showing an improvement of 63% in terms of precision on the selected annotations according to the order that they were presented. The objective is to provide a higher assertion number between suggested annotations and those chosen by the user to annotate a resource.

抄録全体を表示

Recently, we can easily obtain quite a few movies, therefore, we need a method to effectively manage those movies. At this moment, it is difficult for us to find a particular scene in a movie. If we want to do so, we have to watch through all the parts of a movie, we try to retrieve a scene where a human interacts with objects in a room. To achieve this, we firstly apply image processing techniques to analyze a movie and identify human motions in the movie, and then verbalize human behaviors in a room by observing how the human interaction with obects with a model of Bayesian network for identifying human actions.

抄録全体を表示

A neural network type knowledge extraction method from Web and its application to a question-answering system are proposed. In the proposed method, word groups of similar concepts locate in the same layer in the network. The idea is based on neuroscientific findings of Damasio's and a cognitive scientific theory of McClelland's. We carried out two kinds of experiments for the implemented question-answering system and obtained satisfactory results.

抄録全体を表示

Intelligent container system is the state-of-the-art container system which prevents danger beforehand by monitoring status of freight inside the container in real time, and warns the danger immediately when danger is detected. Network device installed in each container is composed of low-power MCU and RF chip, and delivers information of container in accordance with the IEEE 802.15.4. It has the advantage of operating on low-power, but there is a limit in transmitting data due to small transmission frame, and transmission rate falls down in case of being used in a Wi-Fi environment of port. In order to solve this problem, this paper suggests network device which operates with low current by using TinyOS, and uses the communication method selected between Wi-Fi communication and IEEE 802.15.4 communication according to the environment of communication.

抄録全体を表示

As the World Wide Web rapidly grows, a huge number of online documents are easily accessible on the Web. We obtain a huge number of review documents that include user's opinions for products. To summarize the opinions is one of the hottest topics in natural language processing. We focus on aspects of a product in the summarization process. First, we identify a relation between aspects and each word in review documents. Our method employs an unsupervised approach for the identification process. The experimental result shows the effectiveness of the method. Next, we generate a summary by using the relations. In our system, users obtain the summary by an interactive approach.

抄録全体を表示

A twitter user posts a 140-character-limited text named 'Tweet'. So twitter is called micro-blog. In May of 2010, users have posted 15 billion tweets since the service launched. But, tweets are not like web text, it is more personal, temporary, and short contented. Thus twitter's attributes make it difficult to extract keywords from tweets. In this paper, we introduce topic extraction methods using Probabilistic Latent Semantic Analysis (PLSA), and Latent Dirichlet Allocation (LDA) specified in twitter. We show the novel approach to analyze the contents of Twitter user.

抄録全体を表示

The research of multi-agent systems which autonomous agents are able to learn the cooperative behavior is expected in recent years. In our laboratory, we aim at the group behavior generation of the multi-agents who have excellent autonomous learning ability like human through the social interaction between agents to acquire the coordinative behavior. The sharing of environment state can improve the cooperation ability and changing the environment state to the information among agents will get the better cooperation ability. On this basis, we use the reward redistribution among agents to reinforce the group behavior and propose a construction method of the multi-agent system with an autonomous group creation ability, which is able to strengthen the cooperative behavior of the group, as a social agent.

抄録全体を表示

In order for the robot and human to live together, the flexible understanding ability of human intention is required for the robot. In this research, we aim to realize "interactive Emotion Communication" (IEC) which is a bidirectional communication based on emotional behaviors between human and robot. The purpose of IEC is to raise the personal affinity which the robot gives to the human by interactive emotional behaviors between both of them. IEC consists of three processes: (1) recognizing human emotion, (2) generating robot emotion, and (3) expressing robot emotion. In this research, we mainly developed the process of generating robot emotion. We consider that patterns of emotional behaviors desired to the robot vary from person to person in IEC. In this research, we perform the individual preference analysis for emotional behaviors.

抄録全体を表示

Recently, the research and development of the system which is able to obtain the relaxation effect has been actively performed. Most of the moderns have some stresses and require the request for healing, and the early development of the relaxation system is expected. For example, the system which sounds and images are presented to the user, the relaxation effect is judged from the vital signals of heart beat, pulse rate and so on, and the contents of presentation is tuned to raise its effect has been developed. In this research, we aim to construct the system to adjust the automatic operation to obtain the relaxation effect by presenting the sound to the user, and judging the degree of relaxation based on the brain wave information. In this paper, as an advance research on the construction of this system, we try to analyze the features of the brain wave measured in a different state and to find the index that judges the relaxation degree.

抄録全体を表示

This paper shows an inverted-pendulum mobile robot learns dynamic motion from observation of human player's demonstration. The robot has torso and body links and learns ball kicking. Before learning, the robot observes human demonstration with a camera, extracts human links on the images, estimates link posture trajectories, and starts kicking motion learning following them. A reasonable and feasible procedure of learning from observation for an inverted-pendulum robot is proposed and investigated in this paper.

抄録全体を表示

In human robot interaction, intuitive interface is necessary. A specific interaction device, for instance, a joystick or a teaching pendant, is not usually intuitive and needs trainings for a general user. Instruction by gesture is one of the intuitive interfaces and a potential user does not need any training for showing a gesture. Pointing is one of the simplest gestures. We propose a simple human pointing recognition system for a mobile robot and investigate the validity and usefulness with a real robot. One mobile robot with an upward directed camera recognizes human pointing and navigate itself to the place a user is pointing by simple visual feedback control.

抄録全体を表示

This paper proposes an extended linear fuzzy clustering algorithm for square-shape cluster extraction, in which the FCE algorithm is applied in conjunction with local coordinate rotation. The prototypical squares composed of four sides are handled as if it is a single linear sub-structure rotating data points around the centers of squares. In the extended data space constructed after the rotation phase, each linear sub-structure is extracted by estimating its normal vector of local minor component vector. In a similar way with FCE, the objective function is given by a linear combination of the FCM (point-prototype model) criterion and the FCL (line-prototype model) criterion so that the algorithm can perform prototype extraction in a local region. By changing the priority weight of two criteria, the proposed algorithm can estimate not only square-shape clusters but also grid-like clusters having long tails.

抄録全体を表示

Support Vector Machines (SVMs) deliver state-of-the-art performance in many real-world applications such as text categorization, hand-written character recognition, image classification, biosequences analysis, etc. SVMs are one of the standard tools for machine learning and data mining. The classification performances of our proposed fuzzy c-means based classifier (FCMC) on relatively small-sized data sets have been reported. This paper reports the experimental results on large-sized data sets. We compare FCMC with LibSVM by Chang and Lin, which is one of the superb approaches to the SVM classifier for large-sized data sets.

抄録全体を表示

Shell clustering methods partition data sets into several shell-shape clusters by extracting local circles or ellipses as prototypes of clusters. This paper proposes hard c-regression models (HCRMs) for shell clustering. HCRMs successfully detect ellipses by using the hard clustering approach and random initialization.

抄録全体を表示

FCM-type linear fuzzy clustering performs FCM clustering by replacing the point prototypes (mean vectors) with linear prototypes such as lines and linear varieties, and achieves local PCA for extracting intrinsic local linear sub-structures. This paper proposes an extended linear fuzzy clustering model for extracting cross-shape clusters formed by two lines. In the new FCM-type clustering model, each crossing prototype composed of two lines is recognized in a single process by considering local coordinate rotation. After rotation around the intersection of two lines, cross-shape prototype identification is achieved in a similar procedure with the conventional linear fuzzy clustering models where the goal is to identify a single basis vector for each prototype.

抄録全体を表示

Handling relational data is an active topic in fuzzy clustering. In our previous work, we proposed an extended version of linear fuzzy clustering based on Fuzzy c-Medoids (FCMdd), which is used with non-Euclidean relational data. In order to handle non-Euclidean data, beta-spread transformation of relational data matrices used in NERF (non-Euclidean-type Fuzzy c-Means) was applied before FCMdd-type linear cluster extraction. In this paper, the linear fuzzy clustering model is further improved so that the beta-spread transformation is automatically achieved by avoiding negative values for clustering criteria of distances between objects and linear prototypes. In a graded approach, the shift value in spreading transformation is gradually increased considering triangle inequalities among distance measures.

抄録全体を表示

Force and torque induced by traveling motion of a mobile robot effect dynamically to the objects being carried on it. If the induced force and torque should be bigger than the static friction force and torque exerting between the carrying objects and the mobile robot, the carrying objects start to slip. Since this slipping motion causes increasing the acceleration of the mobile robot, then the slipping of one object leads to dangerous collapse of all carrying objects. Furthermore it interferes with accurate traveling motions. On the other hand, mobile robots are desired to transfer the carrying objects as fast as it can. On this view point of contradicted requirements, this paper purposes designing a controller to guide the mobile robot along a given course as fast as possible with acceleration restriction to avoid slipping of carrying objects during traveling.

抄録全体を表示

In a 3D environment, when a 2D figure is photographed from different camera angles, the pictures have a relation that approximates an affine transformation. Detecting an affine transformation is an important problem in the area of computer vision. The log-polar transform is known to be a model of the central-fovea visual sensor in a creature's vision. One study described a method to detect the affine parameter by using a log-polar transform; however, this transform is not invariant to object position, so it is necessary to find beforehand the corresponding characteristic point. In this study, a new log-polar transform method which is invariant to object position change and affine transform is described. One amplitude feature in the frequency domain, the orientation of the 2D figure is retained but change of position is not retained, so the orientation can be detected regardless of the position of the figure. By combination with edge extraction, the affine parameter can be detected with little effect by a background change.

抄録全体を表示