We will show a more comprehensive proof for criterion values in NIS-Apriori algorithm.

View full abstract

Brain-computer interface (BCI) and brain-machine interface (BMI) technologies have recently entered the research limelight. In many such systems, external computers and machines are controlled by brain activity signals measured using near-infrared spectroscopy (NIRS) or electroencephalograph (EEG) devices. In this paper, we propose a novel boosting algorithm for BCI using a probabilistic data interpolation scheme. In our model, interpolated data is generated around classification errors using a probability distribution function, as opposed to conventional AdaBoost which increases weights corresponding to the misclassified examples. By using the interpolated data, the discriminated boundary is shown to control the external machine effectively. We verify our boosting method with an experiment in which NIRS data is obtained from subjects performing a basic arithmetic task, and discuss the results.

View full abstract

The frequency of fall accidents is one of the most serious problems in construction industry in Japan. To prevent these labor accidents, the measures for the fall had been tighten as the institution of guidelines, such as Guideline for Advanced Guardrail Installation Method etc. However, the rate of fatal accidents from falls is still high in the construction industries. Therefore, in this study, the labor accident database by Labor Office were investigated for visualizing the cause of fall accidents and using the safety education in construction industry, and it was found that fall accidents from scaffolds were the most frequent accidents.

View full abstract

The new classification method was proposed which visualizes a multidimensional data set. A dendrogram was created by the distance calculation among the labels. Based on this dendrogram, the colored classification of the cluster group was carried out on the spherical surface based on the label data. This method can be applied to a variety of data. Here, the following problems were analyzed which are the chain-link problem that the representation is generally made difficult, and a problem of the separation of the layered random number planes. In each problem, the representation position of each representative point was examined.

View full abstract

It is clear that each position and the aptitude of a role have effected a team state in organized sports. Many researches are made from standpoints, such as the physical features and mental features. However, the research that shows individual role and relationship within a team in the game in consideration of both sides is very rare. And, the relevance about the amount of physical features and amount of mental features is not yet clear. So, in this research, we aim to develop the clustering method in considering physical and mental features for judging a individual role and adaptation in the game. In this paper, we report the concept of the algorithm and result of the data analysis.

View full abstract

Typical fuzzy reinforcement learning algorithms are based on value-function approaches such as fuzzy Q-learning in MDPs and constant or linear functions are used in the conclusion parts of fuzzy rules. In this paper, we propose a reinforcement learning algorithm based on policy-function approaches where fuzzy functions are used in the conclusion parts as a policy function of an agent.

View full abstract

The cluster distinction in the liver function data was carried out using spherical SOM. The result was also reconfirmed in a plane SOM. There are 5 items of LDH, gamma-GTP, AST, ALT, and ASP, to the main blood examination of the hepatic function which are checked by the checkup. In the analysis by a spherical SOM, the trend of AST resembles very well to ALT. The both standard values were used respectively, for the normalization. Then, they are added and averaged. As a result, it becomes an examination of the hepatic function by these 4 items. Then, the method were applied to the various liver function data.

View full abstract

The acceptance standard of the MS which is recommended at present is hereinafter. It is judged as a MS when the abdominal circumscript exceeds a standard in the first stage and then, 2 items exceed a standard value from the following 5 items of the fasting blood sugar level, the systolic blood pressure, the diastolic blood pressure, the acylglycerol, and HDL cholesterol. In the method, every single item is compared with the standard value. It is then judged to be extraordinary even if when exceeding a little. Also, in the non- pycnic type, it isn't evaluating accurately a MS modality except the abdominal circumscript. We establish a non-ill area outside the standard value. The general MS distinction method was contrived which depends on SOM and the problem of the existing method was verified by this SOM method. by this SOM method.

View full abstract

For learning in continuous value space, reinforcement learning used neural networks is investigated actively. But network configurations of conventional neural network models are fixed, and it is almost impossible to predict and determine proper configuration for best learning performance, because, in reinforcement learning, quantities of training data to obtain enough learning performance are unknown. In this study, we propose a reinforcement learning introduced neuron addition mechanism which is based on RCE model, and evaluate its basical performance.

View full abstract

In this research, model-based reinforcement learning was applied to the six-legged robot. The objective for this robot was to acquire efficient walk movement by model-based reinforcement learning. By setting several constraint conditions on motors of this robot, we considered the reinforcement learning problem with three-dimensional state and action space. Through experiments, the effectiveness of our method using value iteration was revealed in comparison with Q-learning.

View full abstract

Competitive facility location problem (CFLP) is one of optimal location problems for commercial facilities, e.g. shops and stores, and the objective of a decision maker for the CFLP is to maximize the obtaining demands from customers of facilities. This study focuses on a CFLP with uncertain and vagueness demands. By representing the demand for facilities with fuzzy random numbers, a CFLP can be formulated as a fuzzy random programming problem. For solving the formulated CFLP, it is reformulated to the problem as the problem to find their optimal solutions maximizing a degree of necessity under some chance constraint. After showing that one of their optimal solutions can be found by solving 0-1 programming problems, their solution method is proposed by improving the tabu search algorithm with strategic oscillation. The efficiency of the proposed method is shown by the results of applying it to numerical examples of CFLP.

View full abstract

The clonal selection principle explains the basic features of an adaptive immune response to a antigenic stimulus, according to the idea that only those cells that recognize the antigens are selected to proliferate and differentiate. This paper explains a computational implementation of the clonal selection principle that explicitly takes into account the affinity maturation of the immune response. Especially, the immunological memory cells are generated to record the specific antibodies by using clustering results.

View full abstract

We have developed Hiroshima Tourist information Android application called ``Hiroshima Kankou Map'' as smartphone based participatory sensing system. More than 500 subjective data are analyzed by interactive GHSOM and C4.5, and then knowledge related to Hiroshima sightseeing spots are extracted. In this paper, the Android application was improved to send the subjective data with Twitter to construct a mobile social community of Hiroshima tourist. The filtering rules to be tweeted can select the effective tourist information. Moreover, the developed system can retrieve the places where Hiroshima Quest introduces. In this paper, the subjective data from Hiroshima Kankou Map and the tourist information from Hiroshima Quest are analyzed to find a new filtering rule.

View full abstract

In this paper, some imputation strategies are compared in the point that the block diagonal part of the augmented dissimilarity matrix must be filled in for FNM-based and RFCM-based fuzzy co-clustering by entropy regularization, By numerical experiment, the eRFCM-based method with the minimax version of the strategy of the triangle inequality-based approximation and with higher fuzzifier parameter setting achieves the highest value of the normalized mutual information than others.

View full abstract

Recently, school-age children may meet to dangerous situation such as a kidnapping and a natural disaster. In order to keep safety children from these accident, their parents and the social community used to always monitor or shield from danger or harm. However, these countermeasure has concern, because tight protection harms children's growth. It seems to avoid nurturing his/her talents. School-age children has to judge themselves situation in human social community. During school age, they can learn 'bushcraft' through experiences in any natural environment. This paper propose android application which provide kids protect function such as 'kids keitai'. Children's behavior in playground assume that pursuit problem by multi-agent environment. And simulate by hierarchical modular reinforcement uses Profit-Sharing works to plans a target position in higher-layer, and Q-Learning works to trains state-action pair in lower-layer .In this paper, we developed Android application equipped with agent sense possible danger based on rules about children's action come out of learning and emerge children's safety action trigger.

View full abstract

The 2012 Mario AI Competition runs in association with several major international conferences focusing on computational intelligence and games. In Japan, FSS2012 has a special session. The competition of this session focuses on the Learning track. Our team anticipates in a session to verify the effectiveness of our proposed the hybrid search algorithm of game tree, which realizes the pruning by using the reinforcement learning method.

View full abstract

In general, it is not easy to put Reinforcement Learning into practical use. Our approach mainly deals with the problem of designing state and action spaces. Recently, we have proposed a co-construction method of state and action spaces. In this paper, we propose a detection method of environmental changes and a co-construction method by partially enlarging of the calibration of state and action spaces to adapt dynamic environment. In addition, the validity of the proposed method is confirmed through computational experiments using a so-called "path plannning problem" under dynamic environment.

View full abstract

Tactile graphics are useful for visually impaired students when they study mathematics and science. However, producing tactile graphics are not simple task. An intelligent computer-aided system for assisting the production of tactile graphics is needed. To develop such a system, research on mathematical graph recognition techniques is needed. This paper focuses on a method of mathematical graph recognition. A mathematical graph includes graph elements such as rays representing the x-axis and the y-axis, straight line segments or curves representing functions or equations. These graph elements are often overlapped to each other. This paper discusses a method of extracting and classifying such graph elements from the overlapped graph element.

View full abstract

Tactile graphics are useful for visually impaired students when they study mathematics and science. However, producing tactile graphics are not simple task. An intelligent computer-aided system for assisting the production of tactile graphics is needed. To develop such a system, research on mathematical graph recognition techniques is needed. This paper focuses on a method of mathematical graph recognition. When drawing a mathematical graph, we use not only solid lines, but also use broken lines. This paper describes a method for extracting broken line elements from a mathematical graph.

View full abstract

In this paper, we get an inference solutions for natural language propositions involving fuzzy quantifiers and truth qualifiers in case of modifying fuzzy subjects by using genetic algorithms. We apply shift type transformation for fuzzy subjects. We find the inference solutions depend on high level gene number h. The h gets bigger, the tumor of the solutions becomes larger. We use both straight line subjects and S-function subjects. We find no difference between the solutions in straight line subjects and those in S-function subjects.

View full abstract

Ariticulated steering is a system which has front and rear body, connected them at center pin, and turns by the difference between angle of their body. Compared with front wheel steering and rear wheel steering, the turning radius of this type vehicle is small, and both front and rear wheel follow the same path. Such features enable this vehicle is suitable for working in construction. But this steering system is difficult to control the position and angle of the vehicle, because when we change the steering angle, the position and angle of the vehicle also changes. In this study, we propose automated driving system that consider the characteristics of articulated steering type vehicle.

View full abstract

In this paper, "Simple time change fuzzy sets" is proposed aiming to give the appropriate support information under dynamic environment. The proposed fuzzy sets have four auxiliary sets, which enables the operation support system to generate a safe support instruction. It is applied to the lane change support system to let computer determine the proper operation as an human expert.

View full abstract

We would discuss methods for three-way approaches of INDCLUS and INDSCAL into Spatio-Temporal GIS data. Then we applied some effective dynamic graphical approach with two methods to make the spatio-temporal structural analysis for multivariate drinking sales data for safety problem. The last aspect would be dealt with using the dynamic graphics with spatio-temporal presentation.

View full abstract

There exists vast amount of web pages created in the past and left unchanged on the Web. It is expected those pages could be utilized as a resource to know the past. This paper focuses on trend information in the past, and defines context search as a search conducted for knowing the past trend. The context search method proposed in this paper treats retrieval volume by Google Insights for search and search engine's hit count as Web context information. Effectiveness of the proposed method is shown with experimental result, in which objects are game titles.

View full abstract

We propose a classification method for high dimension low sample-size data. High dimension low sample-size data (HDLSS data) is data in which the number of dimensions is much larger than the number of objects. Such data is commonly found amongst real world data such as microarray data or image data. However, it is well known that we tend to obtain poor classification results when applying conventional clustering methods which are based on distance for this type of data. In this paper, we propose an adaptable feature selection method and a related clustering method along with the results of several numerical examples to show the improved performance.

View full abstract

We have proposed a method by using the SOM for determination of membership functions in temperature fuzzy control system. The controlled system is analogically approximated with an electrical RLC circuit and the behavior of the circuit was simulated with mathematical technique under two conditions: non-disturbance and disturbances. The criteria of performance of the circuit were mapped on a SOM map and optimal parameters of the membership function were estimated from the SOM map.

View full abstract

This paper proposes a recommender system based on personal values. Marketing fields have been taking notice of personal values, because such values are significantly related to user preference. While existing recommender systems usually employ user preference of items to make recommendations, proposed method focuses on users' value judgments, on which attribute a user puts a high priority. Implementation of the recommender system based on personal values is supposed to realize the acquisition of diversified recommended items from a non-traditional viewpoint.

View full abstract

Recently,remote operation support systems become important to manage refuse incineration plants. It is used to save energy,to save manpower and to operate in safety. The FCM (Fuzzy c-Means) classifier is a key technology to realize the above systems for substituting operators' vision, because the operators' visual information is very important to operate the plant properly. In this paper, we describe two applications of FCM classifier to image processing and classification for refuse incineration plants. Our methods are more effective than other methods in recognition rate of images. For the future works, we are planning to apply FCM classifier to various fields of refuse incineration plants.

View full abstract

The purpose of this investigation is to examine the effect of education and clinical-training experience about the recognition of personal communication items to the patients of physical therapists and occupational therapists. Subjects were 50 freshmen and 40 seniors in the university who major in physical therapy. The subjects were asked to evaluate these 144 items to see to what degree they would be able to perform adequately and skillfully (i.e. ability score). These items were divided into 4 item groups (fundamental personal skills, specific patient, management difficult and remaining item group). As for all 4 item groups, the average ability scores of the seniors group were higher than those of freshmen group. Various influences on the recognition of the communication skill items of education and clinical-training experience about physical therapy were discussed.

View full abstract

The specification of blood vessel region in the retinal fundus images is one of the important basic technologies in the development of the computer-aided diagnosis systems for funduscopy. In this report, a new blood vessel region extraction method based on the black-top-hat transformation (BTH), which is one of morphological operations, is proposed. In the proposed method, firstly, BTHs with two structuring elements having different radii are applied to the retinal fundus image in parallel. Then the binarization is achieved for the transformed images and tentative blood vessel regions are extracted by using binarized images. Consequently, the final blood regions are determined by correcting the images, which are obtained after synthesizing complementary tentative images. In this study, the experiments using some practical retinal fundus images were executed, and the effectiveness of the proposed method was verified by detection rate of the blood vessels. From the experimental results, it can be said that the proposed method is effective for the blood vessel region extraction from the retinal fundus images.

View full abstract

In this paper, we propose a method to allocate membership functions of Takagi-Sugeno fuzzy model using Particle Swarm Optimization (PSO), which is applied to coronary plaque boundary extraction in Intravascular Ultrasound (IVUS) image. In the proposed method, a plaque boundary is interpolated by the polynomials inferred by the fuzzy rules. The membership functions in the antecedent parts of the fuzzy rules are automatically allocated by PSO. The proposed method improved the accuracy of plaque boundary extraction.

View full abstract

Human can master various motions from training. It usually takes a lot of time and cost. It is well-known that teaching is useful in mastering motions. However, most of teachings tend to teach depending on coach's subjectivity or experiences. The purpose of our research is to establish a method to analyze human's behavior using singular value decomposition and for motion teaching based on its results. Behavior changes through the training, so that motion change point arrays extracted using singular spectrum transformation differ from proficiency. In this paper, a method to identify difference of change point arrays extracted using singular spectrum transformation. Experimental results in mastering and teaching throwing of lacrosse shows the effectiveness of the proposed method.

View full abstract

Genetic algorithm (GA) is an approximate solution method supposed that a network exists. In this research, focusing on integer programming problems, we improve the performance of GA incorporating complex networks. Furthermore, we show the efficiency of the proposed GA method by comparing it with an existing method through the application of them into the numerical examples.

View full abstract

We developed the system that doctors can prevent forgotten medication of patients at any time to check the medication status. This system consists of cloud application on Google App Engine and smartphone application on Android. Smartphone application notifies the time of taking medicine by alarms. The patients can record the medication information by pressing the button. Medication information is sent automatically to the cloud application. The doctors are able to check the medication information at any time using cloud apps in the browser.

View full abstract

In a centipede game, players may make decision based on multiple criteria, not only amount of payoff of the game, by trial and error mechanism. This paper conducts agent-based simulation analysis by using neural network and genetic algorithms. In simulation experiments, information amounts, long-term view about cumulative payoff, and asymmetric agents and their interaction are considered.

View full abstract

We also performed the comparative research in evaluation terms related to the visual user friendless or actual user friendless of a product after the case study of PC mice. Specifically, we evaluated the paired comparison about each evaluation term using six kinds of electric razors and asked for the ranking by Interval AHP. We conducted cluster analysis based on the ranking, and took the comparison consideration including the case study above.Then we found out the different between case studies.

View full abstract

In a robot production subject at universities and high schools, robot competition is employed for aiming to motivate students and to encourage intelligibility of technology and knowledge. Robot competitions have a high degree of availability for reducing problem of science course and engineering. In this presentation, we introduce the basic robot production subject at department of electronics and mechatronics, tokyo polytechnic university.

View full abstract

This paper presents a Takagi-Sugeno fuzzy modeling and linear matrix inequaility (LMI) based parallel distributed compensation (PDC) approach to modeling and control of an F16 aircraft nonlinear model. After presenting dynamics of an F16 nonlinear model, fuzzy model-based control results are discussed in comparison with linear control results. The fuzzy controllers designed using the LMI based PDC approach guarantee not only stability but also decay rate and constraints on control inputs (actuator saturation) and outputs. Finally, simulation results demonstrate that the fuzzy controllers can stabilize the original (not simplified) nonlinear F16 model with given initial conditions although the linear controllers cannot stabilize the systems.

View full abstract

Emotion estimation from sentences is helpful in semantic analysis and human-computer interaction. People sometimes express their emotions directly and sometimes describe experiences that produce various emotions. Emotion estimation requires both verbal expressions. We believe Japanese onomatopoeias could serve as a landmark of emotions. In this study, multiple raters judged emotions found in 324 onomatopoeias. %We classified 324 onomatopoeias into eight emotional categories based on the evaluation. We quantitatively evaluated the degree of agreement among the raters and attempted to improve it.

View full abstract

Inexact information such as human behavior and cognition could be properly analyzed and clarified by applying fuzzy theory. In this paper, the authors present an analysis method of the opinion survey by applying fuzzy graph and also illustrate its practical effectiveness with the case study concerning the subject preference of high school students.

View full abstract

In our study, prediction of the number of asthmatic attacks in Himeji city next month from 1 to 4 years old and 65 years old or older. Fuzzy-AR model learns from 2001 to 2005 and predicts the number of asthmatic attacks using data from 2006 to 2010. We evaluated our prediction results by using correlation coefficient between predicted value and ground truth value and compared Fuzzy-AR model and AR model. As a result, 65 years old or older was not able to be predicted by each model. However in age of 1 to 4, Fuzzy-AR model was able to predict higher correlation coefficient than AR model.

View full abstract

This paper proposes a context search system for exchange rate data,which supports users to grasp the relationship exchange rate and events. As the trend of exchange rate is influenced by various kinds of events, it is important to grasp how a certain type of events will influence the trend of exchange rate, and the type of events corresponding to time periods having similar trends. This paper defines a context search as the function of searching similar time period having similar trends with a specified time period, and that of summarizing events occurred in the time period. This paper examines the proposed functions based on the prototype system.

View full abstract

AR model is a model that is applied to time-series data in statistics, is used to predict the future from past data. In this study, we predicted using the AR model the number of asthmatic attacks of each region in Himeji, investigated the relevance of the region and cause of asthmatic attacks, and aimed to improve the accuracy of prediction system the number of asthmatic attacks by age. The result is estimated the number of asthmatic attacks in 2006-2010 from data 2001-2005, were compared with the true value. In a particular region, good results was obtained in the prediction of the elderly had been considered a low autocorrelation. Therefore we confirmed that asthmatic attacks involved not only generation but also region.

View full abstract

There have been many studies on semi-supervised classification, but methodological aspects such as generalization capability of a particular method and relation of solutions by different methods of clustering is relatively unknown. We consider a family of method related to K-means, and show how the above problems are studied and solved. Moreover methods of on-line K-means are generalized and semi-supervision is incorporated.

View full abstract

This paper deals with a design of the fuzzy model following control systems with disturbances. We guarantee the boundedness of the inner states for the control systems. Numerical example shows that the output signal of the control systems asymptotically follows the reference model output signal with disturbances.

View full abstract

In the field of robot-assisted therapy, the impression evaluation for robot behaviors are important for enhancing the effect of treatment.This paper presents the impression evaluation for mixed walking motion of animal type quadrupedal robot "AIBO" .We verified the effects of different behaviors on human sensibility evaluation by comparing between the normal pre-installed gait and the animal-like gait generated by imitating canine walking, optimizing a leg path using a genetic algorithm and applying a zoological basis.Moreover, we analyzed the relationship between behaviors and impressions by evaluating the three mixed walking motions.

View full abstract

The participant's attribute and purpose differ from each other in each robot contest. Howeber, various capability required of a robot maker is common. Those human skills are important for an especially young engineer.

View full abstract

We propose Chain Form Reinforcement Learning for a reinforcement learning agent. In the real world, learning is difficult because there are an infinite number of states and actions that need a large number of stored memories and learning times. To solve a problem, estimated values are categorized as"GOOD" or "NO GOOD" in the reinforcement learning process. Additionally, the alignment sequence of estimated values is changed because they are regardedas an important sequence themselves. We conducted some simulations and observed the influence of our methods.

View full abstract

Reinforcement learning has important aspects for exploratory try-and-error and delayed reward. However, it needs many learning cycles and takes a long time to get a sufficient reliable learned result. In order to solve the problem, there is a transfer learning, which requires how to decide the transferring knowledge and how to transfer it. Additionally, an achieved knowledge is not always suited for the task in the real world. In this research, we proposed a communication method based on PCA and discussed these problems with several methods and simple simulation.

View full abstract

Army ants performs the altruism, in which an ant sacrifices its own well-being for the benefit of another ants. We developed the multi agent system of army ants. A kind of altruism is appeared in the simulation results of the system which has 2 kinds of ant agents, `Major ant' and `Minor ant'. Ants communicate with each other pheromones. The pheromones evaporate at a ratio and diffused into the space of neighbors. The other ants can recognize them as the signals from the other ants. In this paper, the distribution of diffusion is performed on a stochastic function. This paper describes the performing altruism of ant based multi agent system.

View full abstract