With the rise in popularity of artificial intelligence, the technology of verbal communication between man and machine has received an increasing amount of attention, but generating a good conversation remains a difficult task. The key factor in human-machine conversation is Whether the machine can give consistent emotional expression as a response. we propose a model to estimate emotion by considering time series by autoregressive model. This model extracts emotional values from sentences and makes consistent emotion estimation by time series processing of emotional values. We perform a comparison between our models and comparative models and find that we can get a slightly better result with respect to emotional consistency. Problems also exist in emotion estimation at the current research stage, but time series processing proved to be effective in emotion estimation.

In this paper, we propose a decision tree construction method to complement missing evaluation items from questionnaire items in Kansei analysis method with Fuzzy C4.5 decision tree. When questionnaires to be carried out for impression analysis omit items that users emphasize, Kansei analysis might not be performed appropriately. Therefore, we generate a decision tree using the Fuzzy C4.5 algorithm and estimate the evaluation value of the missing evaluation items from the relation of the information gain ratio which is the feature of the algorithm. Next, we compare the classification abilities of Fuzzy C4.5 decision tree that complement the estimated missing items with standard one through simulation.

Increasing motivation of students in mathematics subject classes is strongly linked to academic achievement, and teachers are required to have educational skills to motivates students. Indeed, modern pedagogy points out that motivation of students themselves has a great influence on learning effect. Also, improvement of student motivation is very important. In this paper, we would conduct a questionnaire to grasp students' motivation and investigate each factor by using portfolio analysis. Furthermore, we try to extract improvement items and their priorities in the lesson.

Probabilistic topic models based on latent Dirichlet Allocation is widely used to extract latent topics from document collections. In recent years, a number of extended topic models have been proposed, especially Gaussian LDA(G-LDA) has attracted a lot of attention.G-LDA integrates topic modeling with word embeddings by replacing discrete topic distribution over word types with multivariate Gaussian distribution on the word embedding space. This can reflect semantic information into topics. In this paper, we use a G-LDA for our base topic model and apply Stochastic Variational Inference (SVI), an efficient inference algorithm, to estimate topics. This encourages the model to analyze massive document collections, including those arriving in a stream.

A word vector of distributed representation that embeds the semantic relation among words into the vector using Word2Vec has attracted attention in recent years. Furthermore, this word vector has become widely used in the field of Natural Language Processing such as parsing and document classification, and its effectiveness has reported. Generally, only the input side weights on Word2Vec are used as the elements of a word vector, and the output side weights generated at the same time are not used. On the other hand, the authors focus on the usefulness of the paired output side weights with the input side ones. In this paper, we propose a word vector using the input and output side weights together. In addition, we experimentally investigate the performance of the proposed word vector.

Sentiment analysis is a technique that analyzes and extracts the evaluation information for products and services from text data. In the present paper, we propose variations of unsupervised sentiment analysis models that can estimate semantic orientations of comments (whether they are either positive or negative) together with their topics by processing text data without any ratings. The proposed models extend a standard topic model by introducing dictionaries of emotional words so that it can analyze in what point of view people are evaluating subjects positively or negatively.

This paper describes a method for extracting opinions with a basis from assembly minutes of Japanese 47 prefectures. The authors released the web-based system for searching the assembly minutes of 47 prefectures. Users can overview speech records including a search keyword using the system. Furthermore, the system can show regional differences of speech record volume as a color-coded map. On the other hand, it is difficult for the system to extract speech records including opinions because of indetermination of search keywords. Such speech records including opinions are good materials for understanding political concept of assembly members. In this research, we examine a method for extracting sentences including opinions using predicate argument structures.

In this paper, we proposed an exercise promotion system that considers user's preference and aimed to maintain motivation for user's exercise. Nowadays lifestyle diseases are prevalent all over the world and people who are interested in health care are increasing. One way to prevent lifestyle related diseases is to continue exercise. However, continuing exercise is expected to be difficult due to various factors. We considered that it is necessary for the user to feel exercise as fun in order to allow the user to continue exercise. Therefore, we constructed an exercise promotion system using user's preference and conducted experiments as to whether users can feel the enjoyment of exercise. As a result, we were able to make the user feel pleasure of exercise.

Human errors during simple tasks sometimes cause serious accidents. These errors are sometimes caused by mind wandering. In this paper, we focus on mind wandering during simple response tasks. We carried out AX-CPT test with neutral situation and with fidgeting. We investigated the results on error rate and user's fatigue while experiments.

Stress is becoming a very serious problem for people currently living in society. In this research, we propose a method to measure stress daily by using a wearable device, and to obtain a natural measurement result from conventional research. In addition, we propose a series of systems that manage user’s stress from real - time stress measurement data and support the approach to relieve accumulated stress.

Quantitative analysis of human brain activity has been actively studied in brain and neuroscience. In the analysis, there has recently been getting many opportunities that deep leaning is applied to deal with brain activity data observed with fMRI. Whereas, it costs expensive to collect such data and this always causes the lack of enough data to train a model in deep learning framework. With this background, we aim to artificially increase the number of such data and apply them to a caption generation task with deep learning for raising the accuracy of the model.

Human have two types of inferences: intuitive and logical. In traditional inference studies, the intuitive inference has modeled by a probabilistic method like the Bayesian, and the logical inference has modeled by a symbolic method like the Tree search. There are many studies that relate inference behavior with brain areas, but few have focused on the mechanism of logical inference that can emerge naturally from the brain neural circuit. So, in this study, we assumed that the human inference process should be modeled as an operating mode switching of a single distributed neural network. We used an associative memory model for its verification. The intuitive inference function is realized by a combination of a memory association from a current state and an association of the memory state with the value recognition. Then, the logical inference like behavior is realized by repeatedly biasing the gain of the valued memory state found in the intuitive inference process. A computer simulation in a maze search task is conducted, and we confirmed the emergence of the symbolic tree search like inference behavior with pruning of low probability branch from the intuitive like probabilistic inference by the change of the calculation parameter of the same model.

In this paper, we propose an improved Hierarchical Temporal Memory (HTM) that can consider long-term dependence. Conventional HTM is a temporal sequence prediction model imitating the cerebral cortex structure and learning algorithm. In the conventional model, only the connection with the previous data is learned, but in the proposed HTM-TA the connection with several former data can be learned. Therefore, we add the time axis to the segment which is the collection of synapses in the structure of HTM and we name this HTM-TA. As a result of evaluation experiments, it was confirmed that the proposed model obtained higher accuracy than the conventional model on temporal sequence prediction.

Learning type trapezoidal fuzzy inference is a general form of triangular fuzzy inference, and fuzzy clustering using fuzzy inference can identify the class boundary area by learning. It is necessary to adjust the trapezoidal membership function of the antecedent part and the singleton real value of the consequent part with the learning mechanism. In addition, initial value setting of parameters is an important factor for learning for minimizing error. In particular, the initial value of the singleton in the consequent part greatly changes the accuracy of the estimation of fuzzy inference depending on the set value. In this paper, we will discuss the characteristics of the hyperparameters to determine fuzzy rules of class discrimination with minimizing error. In addition, we discuss the influence of the learning parameters of the antecedent part and the consequent part on error accuracy, and the procedure of learning for improving learning. By the numerical examples, several combinations of extensive hyperparameters are discussed by significance test in order to acquire the optimal fuzzy rule of fuzzy clustering.

The present paper proposes a computational model of the human drivers’ vision named “moving object attractiveness map”, aiming at image processing technique to analyze unsafe traffic conditions that may provoke their inappropriate allocation of visual attention. Saliency map is a computational model of image processing to simulate human visual attention after scientific findings of bottom-up information processing at the low-level visual cortex, which extracts distributions of salience features from a given image and combines them to estimate which areas in the image can attract the observer’s visual attention more. The proposed model adds a top-down attention filter to the saliency map that predicts the likelihood of visual objects to attract the observer’s attention for possible collisions.

A pairwise comparison matrix (PCM) is often used to estimate the weights of criteria/alternatives. In the conventional methods, a crisp weights have been estimated from a given PCM although the given PCM is not perfectly consistent. Namely, crisp weights are estimated by minimizing the errors. In the interval analytic hierarchy process (interval AHP), weights are estimated by intervals from the point of view that the human conceives interval weights rather than crisp weights in their mind. Because of the insufficiency of the original interval weight estimation method, several alternative estimation methods have been proposed. For such an interval estimation method, the satisfaction of the following properties are preferred: (1) the estimated interval weight vector is normalized, (2) the given PCM is a possible realization under the estimated interval weight vector, (3) the correct crisp weight vector is estimated from the perfectly consistent matrix, and (4) the more PCMs are observed, the closer to the true interval weight vector the estimated interval weight vector comes. Properties (1) to (3) have treated considerably. On the other hand, property (4) has not yet treated so far. In this paper, we propose several estimation methods having properties (1) to (4).

Interval UTA method has been proposed as an intermediate model between the conventional UTA and UTA-GMS methods. It can be seen as a semi-robust ordinal regression method providing a safer evaluation than the conventional UTA method with a less computation than UTA-GMS method. We found that the preference relation obtained from the identified interval model does not match well to the preference probability between alternatives obtained by a Monte Carlo simulation. We improve the identification method so as to match more to the preference probability by replacing the objective function of the identification problem with another one. We confirm that proposed method is improved by numerical experiments.

Interval UTA method has been proposed as an intermediate model between the conventional UTA and UTA-GMS methods. It can be seen as a semi-robust ordinal regression method providing a safer evaluation than the conventional UTA method with a less computation than UTA-GMS method. We found that the preference relation obtained from the identified interval model does not match well to the preference probability between alternatives obtained by a Monte Carlo simulation. We improve the identification method so as to match more to the preference probability by replacing the objective function of the identification problem with another one. We confirm that proposed method is improved by numerical experiments.

It is important to understand tourists’ behavior in order to promote tourism promotion measures and solve various problems by increasing tourist. Tourist’ behavior survey have been done by questionnaire survey. In recent years, surveys using rent-a-car probes and GPS functions of mobile phones are being conducted. However, both have problems such as staying at a certain period of investigation, costly expenses, and inability to investigate round-trip behavior by walking. Therefore, in this study, we installed Wi-Fi packet sensors at 50 locations in Okinawa Island, and estimated the round - trip behavior. This paper show the feasibility of the estimation results and the applicability of the data and possibility of fusion with other surveys.

The market scale of e-commerce sites is expanding, and the number of e-commerce sites is increasing. In such a situation, it is important to analyze what kind of products and information they are interested to do marketing. It's said that interest in specific products are correlated with reading and reading hours, based on them. However, the contents vary depending on each page, so even if the number of browsing is the same, there are differences in the information that a user has acquired. It is necessary to extract the interest and interest after considering the information collected by the users. Therefore, in this research, we considering the access page and the content of access, we will devise a method to estimate the possibility of purchasing and evaluate the effectiveness.

Under circumstances out of the mountaineering path, climbers will lose sight of their current location or mountain trails when it becomes impossible to acquire location information from GNSS. Along with that, it will lead them to lose their path which is the maximum factor of distress. For deterring distress, it is necessary to gain a highly reliable movement trajectory at the time of straying. However, the location information acquired from GPS sensor of respective smart phones exhibits unevenness in the accuracy and contains plenty of noises since they differ from each other depending on the around environments or models. As a solution to this problem, a noise elimination method of GNSS in a smart phone has been proposed, but it is difficult to acquire continuous position information in mountainous areas where there are many obstacles such as trees. In this research, we developed a system to calculate moving distance and azimuth using other sensors built-in smart phone, and tried to apply it to assisting position positioning.

The Cabinet Office advocates Society 5.0 in the 5th Science and Technology Basic Plan. This is an effort to solve the problems of society by connecting all people and things through the Internet, and highly integrating virtual space and real space. For example, there is a smart kitchen summit on the theme of "food and cooking × technology". At the summit, various kitchenware have been proposed aiming at creating a future kitchen concept and a new lifestyle. However, in order to realize, it is necessary to replace existing kitchen appliances, and realization is unclear in many cases. Therefore, in this research, we propose a method to smart kitchen using range image sensor and small projector. Then, we develop Concept Kitchen 2025 and feasibility is considered.

Cleaning robots such as Roomba are popular in homes and offices in recent years. People sometimes need to support the robot while the robot performs cleaning task due to obstacles which interfere with the task. In this situation, the task can be carried out more smoothly if the robot can communicate its intention about the task to the people and arouse the people’s action to assist the robot. Hearing-dog robot AcToR was previously reported as an robot communicated its intention to a user. AcToR can move and touch the user. The touch behavior achieved to communicate urgent information to the user. This paper proposed to apply the touch behavior to communicate AcToR’s intention to the user and arouse the user’s action to asssist AcToR. Pet dogs and cats sometimes perform actions such as scratching the door when they want the door to be opened. If AcToR can perform similar behaviors to the pet actions, it is expected that AcToR can communicate its intention to the user and arouse the user’s action to asssist AcToR. We conducted an experiment to confirm usefulness of the robot touch behavior. The robot task is to open a door of a room and go out of the room. The experiment was conducted under two conditions “touch to the door” and “no touch”.

A hearing dog is a sort of assistance dog for hearing impaired people. Physical touch of the dog can alert the people to important life sounds, such as doorbells, alarm clocks, or fire alarms. Hearing dogs can work to assist the people, but the working number of hearing dogs are insufficient all over the world. Instead of hearing dogs, we have been developing hearing-dog robots. Our robot can autonomously move to search a user. A stochastic model using past experience about locations where the user stayed was examined and confirmed as efficient to searching the user. This paper proposed new searching algorithm of the robot to consider user’s rhythm of daily life with time information of past user locations. The experimental result indicated.

We are developing the presentation robot system that performs flexible interaction with a group of multiple participants. In order to take a flexible conversation, a robot requires an interaction based on a reaction of a group. In this paper, we develop the environmental measurement system for the robot interaction. We have used a spherical camera, a web camera, acceleration sensors, and a Kinect sensor. As an experiment, we perform the demonstration of the robot presentation, then measure the reaction of participants using our developed system. We discuss the potential to decide robot interaction behavior by using the results of the spherical cameras and acceleration sensors.

In this paper, a method that applies the difference of feature amount in human facial expression learned by SOM to the emotion transition in Markovian Emotional Model and generates an emotion reaction model of a communication robot is proposed. For example, it is thought that what an angry person bursts into tears than laughter occurs with high possibility when the emotion of person changes. The expression getting angry is more likely to resemble a crying expression than a laughing expression at that time. In other words, the emotion which the characteristic of facial expression resembles is easy to change. The emotion transition probability is obtained by a questionnaire, and an interaction experiment using a communication robot is performed in this study.

Recently, SSVEP- based BMI could support various aspects of everyday life of elderly and disabled people. In this research, we developed a noninvasive BMI system that controls the robot hand using induced brain waves SSVEP in order to improve the quality of life(QOL) of patients with hands or arms deficient or impaired. This BMI system consists of visual stimulator, 6 degree of freedom (DOF) robot hand, an EEG recorder and a laptop for processing data. The subject induces the corresponding SSVEP signal by seeing one target in the three visual stimuli (5Hz, 6Hz, 7Hz) representing the motion: grip, pinch, arm rotation of the robot hand, respectively. The detected SSVEP signal is classified by canonical correlation analysis (CCA). The robot hand is operated by converting the SSVEP into the control signal according to the classification result. The results show that the proposed BCI system has a high performance, achieving an accuracy of 93% in a window length of 1 s and the second harmonics.

Recently, there is concern about labor shortage of nursing care markets accompanying a declining birthrate and aging population as a social problem. Power assist at the nursing care site is effective when holding a person, but it takes time to attach and detach. It is not easy to use and inefficient when is viewed through the day's work. There are two types of myoelectric sensors used for power assistance, wet (passive) and dry (active) electrodes. Performance as a sensor used for power assist is better for dry electrodes and less troublesome for attachment and detachment. However, it is hard to use to adhere to the skin for a long time because the stiff case of the sensor may bring about the pain to the user. It is not suitable for power assist in existing hard-casing electrodes. The purpose of this research is to develop an active soft-casing sensor, which does not give the pain for a long time using a power assist.

There is a problem of exposure with X-ray for surgeon and staff in neuroendovascular treatment. We developed a remote operation robot to solve this problem. It can control both a catheter and a guide wire by using two joystick controllers. In this paper, we describe details about developed the robot and its evaluation.

In this research, we developed a scenario-typed robot therapy program with the aim of creating group communication among elderly people with dementia through robot. Specifically, we developed a program to conduct group activities (games) with pet robots after creating an atmosphere of communication through a pet robot. This program consists of touching time with pet robot, gymnastics with pet robot, and game with pet robot. Then, the frequency of occurrence of communication and the expression time (ARS) of emotions of elderly people during conducting game with pet robot were examined. As a result, when the elderly people participate in the game, it is found that the communication amount increases by 10 times or more on average as compared with the normal living hours zone. Furthermore, it was revealed that there is a strong positive correlation between frequency of occurrence of communication and ARS value (time of expressing good emotions).

In LiDAR measurement, measurement loss caused by laser reflection not returning is called black spot. A black spot occur when a laser is irradiated on a specular reflection material or a material with a high absorption rate of infrared rays, in many cases, laser reflections are obtained from around the black spot. Conventionaly, in object recognition using LiDAR measurement, only measured data has been used. However, since LiDAR actively irradiates the laser, we thought that black spots also contained information for object recognition. Therefore, we propose a complement method of black spots in LiDAR measurement Moreover, we confirmed through several experiments that the pedestrian recognition accuracy improves by applying the proposed method.

As a limitation of the vision sensor or camera in the night and bad weather conditions, EHF-band radar is used to compensate the limitation. However, the disadvantage of the EHF-band radar is automatic generation of the clatter and multi-pass waves. In the present study, we applied the Probability Hypothesis Density (PHD) particle filter to treat the problem.

As a deep learning algorithm, software tools based on Generative Adversarial Network (GAN) are released in the form of pix2pix. In the present study, we report a preliminary study of the application of pix2pix to the automatic conversion from aerial photos to vector representation in the digital map and discuss its limitation.

In Otoyo-city of Kochi Prefecture Japan, wild monkeys have appeared farm field at several years ago, and an agricultural crop damage has been occurring. This district is a region mainly based on agriculture. Damage caused by monkeys has a direct influence on the production amount of agricultural crops. Catching by using a trap, requires the detection sensor to activate a trap. In general, the detection sensor is realized by the contacting wires or an infrared sensor. However, it can only detect a single monkey. In the case of monkeys, it is necessary to capture the whole herd. Because other monkeys will not get close to the trap even if capture the single monkey. We develop an artificial intelligence system that detects a group of monkeys and a catching trap system with them. This system use camera and target detection system, send images to a network server. In this server system, we developed an AI system to class monkey images and others. This paper shows the developed system and structure of our system and method.

In this paper, we made line-trace car. We implemented ESP-WROOM-32 with low-price, high-performance WiFi built-in microcomputer and Linescan Camera Module TSL1401-DB.Report the result of running.In the implementation experiment, we tried simple fuzzy control with four Reflective Object Sensor. In addition, line scan images were acquired with TSL1401-DB. We got a smooth running data and also a line scan image with TSL 1401-DB.In the future research we will develop a line trace car using line scan imaging sensor.

Mini 4WD AI is a research competition for cheap and highly useful systems. The aim is to make the mini 4WD run wisely by controlling with AI. In control of mini 4WD, accurately grasping the state of mini 4WD such as the position and the direction is an important factor. There is SLAM as a method for simultaneously creating such an environmental map and estimating its own position. In introducing SLAM, it is important that multiple pieces of information can be correctly judged as information on the same location. Therefore, this research aims at estimating the same location using DP matching.

These days, we have many sensory data network with massive devices so called IoT. These data are too many and complicated. So usual IoT takes time to do tha calculation.Fuzzy logic can calculate many and complex data fastly.In this paper we propose fuzzy modeling to recognize such a massive data heap.

Sighted people use not only verbal information, but also visual information in order to understand the subjects such as mathematics and physics and so on. Although visually impaired people are able to access visual information when the information is expressed by a tactile graphic, it is quite difficult for them to produce figures by themselves. Using refreshable braille display and keyboard is one of the possible methods when visually impaired people draw figures by themselves. So, we will discuss an assistant system for visually impaired people to draw figures by themselves using refreshable braille display and keyboard.

Blind people often difficult understand tactile graphics information in braille text, because tactile graphics drawing complicated. In addition, there is little knowledge about touching tactile graphic and an effective teaching touching method has not been established. In related research, we are not conducting experiments in a situation close to the environment normally used by blind people. It is necessary to make quantitative measurements and do not limit the touch and do experiments with the notation method used in braille text. Therefore, in this paper, in order to acquire knowledge about efficient tactile sensation, this time we will investigate the influence on the tactile figure survey using the tactile graphic feature struggling to grasp figure and the tactile observation method, we investigated discrimination threshold of dot pattern used for graphic representation.

At present, orthopedic surgery has a large variety of surgical techniques. Procedures are complicated, and many types of equipment have been using in the surgery. So, operating room nurses who deliver surgical instruments to surgeon are supposed to be forced to incur a heavy burden. Although there is a navigation system for assisting surgeons in artificial joint replacement surgery, but no system exists for assisting operating room nurses. This work proposes a computer-aided navigation system that indicates the current procedure and procedure progress for nurses, and also instructs nurses to prepare surgical instruments to be used in the next procedure using smart glasses. Firstly, the system estimates the current status of the surgery procedure using a convolutional neural network (CNN) by utilizing realtime video images taken from smart glasses which was worn by operating surgeon. Then, the system indicates nurses the surgical instrument to be used at the next procedure in the smart glass worn by the nurses. The system was implemented with the object detection technology and the augmented reality. Experiment results demonstrated a satisfactory performance of our proposed system of recognizing surgery procedures and detecting surgical instrument.

As soft computing extensions of Hard C-Means (HCM) clustering, Rough C-Means (RCM) and Rough Set C-Means (RSCM), which can deal with positive and possible cluster memberships based on rough set theory, have been proposed and utilized for detecting vague boundaries among clusters. Semi-supervised clustering schemes that utilize not only unlabeled objects but also partial labeled objects are promising approaches for improving the classification performance. In this study, we consider how to introduce semi-supervised approaches to RSCM clustering. Furthermore, we confirm the effectiveness of the proposed method through numerical experiments.

In the field of fuzzy clustering, possibilistic clustering approaches that generate possibilistic partitions by eliminating probabilistic constraints have been reported to be effective for analyzing noisy data. Additionally, graded possibilistic approaches that gradually relax the probabilistic constraints and realize transition between probabilistic and possibilistic partition have been proposed for utilizing advantages of probabilistic and possibilistic clustering. Fuzzy co-clustering is a promising approach for analyzing cooccurrence information. In this paper, we propose a graded possibilistic approach for fuzzy co-clustering that realizes transition between probabilistic and possibilistic partition. Furthermore, we confirm characteristics and performances of the proposed method through numerical experiments.

In order to improve the interpretability of fuzzy co-cluster structures, some modified models have been proposed for introducing exclusive feature not only into object partition but also into item partition. This paper proposed a novel modification of multinomial mixture models (MMMs)-induced fuzzy co-clustering, where Ruspini's condition is forced to both object and item partitions. Through numerical experiment with artificial data and application to document analysis, the characteristics of the proposed method are compared with the conventional model, which achieved exclusive partition of items by introducing a penalty term.

In this paper, we focus on the granularity that indicates the level of fineness of features obtained from data. We define the granularity of features as the number of clusters formed by a batch of those feature vectors, and analyze the effect of granularity distribution of features on machine learning tasks. For this purpose, we propose a new neural network architecture and the regularization method that can control the granularity of features explicitly. The proposed network has a branched structure inside and learns multiple feature subvectors regularized to be evenly divided into the specified number of clusters. In experiments with classification task, we achieved higher accuracy than the usual one when we learn feature vectors in an unsupervised manner. We also confirmed that the existence ratio of coarse-grained features leads to better representation for classification rather than the multi-granularity property.

Goal of our work is to develop a nondestructive evaluation support system for fish meat using ultrasound. The reason for using ultrasound is that not only can noninvasively and quantitatively evaluate fish in alive but also be expected to be able to directly evaluate fish swimming in the water. In previous works, we have proposed an evaluation method using the integrated backscatter (IB) method applied in the intra-vascular ultrasound. Moreover, we have shown that the it is possible to quantify evaluate the fat content and the texture of the fish meat by using the IB method. However, the sufficient evaluation accuracy for the practical applications has not been obtained with the IB method. In this work, we propose a new nondestructive fish meat evaluating method composed of the self-organizing map and the radial basis function network. As a result of verifying the usefulness of the proposed method through experiments, the fat content and the texture of fish meat were estimated with high accuracy.

Unlike classical statistical treatment, Tokutaka et al. proposed a significance by SOM based on a new concept that ”significance is a magnitude relation between two data class elements”. There are several applications of the proposed method by Tokutaka et al., However, since the proposed method is a new concept, it is not yet enough to elucidate basic characteristics.Therefore, using the iris benchmark data of Fisher-Anderson used in data classification, the properties of significance are clarified, and its effectiveness is clarified from application to fundus image analysis as a practical application.

In this paper, we introduce trials on extraction and processing method of brain wave (EEG) waveform when used for analysis of self-organizing map etc. We propose extracting waveforms using ECG, which is often measured simultaneously with brain wave measurement. We report the analysis results of extraction and processing of brain waves measured during resting and light exercise.

When SOM learning algorithm is applied to a large amount of data, it can be an effective approach that a parallel and distributed processing is adopted. In this study, we have proposed a kind of implementation method which parallelize the SOM learning algorithm by dividing the learning dataset. Furthermore, GPGPU approach is also adopted for vector operations under SOM learning process and we have confirmed an effectiveness of these implementation methods with respect to a computation time. In this paper, we propose another kind of implementation of parallelized SOM which divide competitive layer of a large scale SOM.

In road projects, the Administrators have drawn the Business Continuity Plan (BCP) on the assumption that a large-scale disaster occurs. And based on this robust disaster prevention system, they have been working on maintenance and management of roads every day. To improve their work, we focus on SNS which has latent information can be used for grasping road’s status when a disaster occurs. In this research, in order to complement and expand existed information collecting systems owned by road administrators, we report the result of speedily collecting and analyzing technologies for the road information like disaster from SNS. In this paper, we conclude all results obtained from studies in the 3-years-plan targeting Hanshin expressway.

In recent years, road managers have been using probe data to understand the actual condition of car traffic. The Ministry of Land, Infrastructure, Transport and Tourism promotes the use of ETC2.0 probe data, and the number of vehicles equipped with ETC 2.0 currently exceeds 3.7 million. The amount of accumulated data of ETC 2.0 probe data, which is the traveling and behavior history of a vehicle, has increased dramatically. Based on the amount of data accumulated and acquisition characteristics, we consider understanding the actual condition of car traffic from both the viewpoints of macro and micro to be possible. ETC2.0 probe data can be analyzed in more detail than the data used for existing road traffic analysis. In this study, the authors propose a road traffic analysis method that takes advantage of the unique features of ETC2.0 probe data.