Clustering is a method of data analysis without the use of supervised data. Even-sized clustering based on optimization (ECBO) is a clustering algorithm that focuses on cluster size with the constraints that cluster sizes must be the same. However, ECBO has the problem that it is susceptible to noise. It is believed that this issue can be overcome by applying noise clustering method. Noise clustering is a method that makes noise less susceptible to noise by classifying it into noise clusters by identifying noise. In this paper, we propose a new even-sized clustering algorithm based on noise clustering and verify its effectiveness through numerical examples.

The c-regression is a clustering algorithm that performs regression analysis at the same time by using the dataset of independent variables and dependent variables. Therefore, the c-regression can classify data into each tendency of data. Incidentally, when clustering is used in the actual society, there is a situation where it is preferable that the number of data to be classified is uniform. This paper proposes a new c-regression with constraints on cluster size by introducing size equalization concept to the c-regression.

Entropy-regularized Fuzzy c-means is a representative fuzzy clustering algorithm, and its properties have been investigated in detail. In particular, it was proved that the fuzzy classification function value at infinity is crisp, i.e., zero or one under a certain condition. In this report, under the case that the above condition is not satisfied, the fuzzy classification function value at infinity is discussed.

Growth of the Someiyoshino cherry trees has peaked from 40 to 60 years after planting and the trees tend to decline rapidly later and curing such as fertilization at this peak has a remarkable delayed effect of decline of the trees. It is the most important task to grasp the health condition of the trees. We collected the data of cherry trees and old trees of cherry blossoms in parks, schools, riverbanks, etc. in various places, and investigated distribution status and characteristics by using the spherical SOM.

Air pollution is a major theme of environmental problems in recent years. Health hazards to respiratory organs by pollutant microparticulate matter (PM 2.5) and photochemical oxidant have been reported. Since necessity of disclosure of air pollutant concentration information and prediction of atmospheric pollutant concentration is increasing in our lives. The current prediction method by numerical model has a problem that require enormous research and advanced scientific knowledge and mathematical knowledge. Although there is accuracy that is qualitatively predicted on the East Asia scale, accuracy of city scale and quantitative accuracy are insufficient. Therefore, We verified the prediction using self-organizing map as local governments etc. can predict without advanced knowledge and introduction of new funds.

　In recent years, in the Koyama-Lake in Tottori, the outbreak of the red tide becomes the problem that a sea mingled with fresh water.As a means to grasp the causal relation of the water quality condition of a lake, there is a method using a numerical model, but water quality simulation using numerical models requires understanding of factors that govern phenomena to be studied, and good results cannot be obtained unless an appropriate model is selected according to flow / water quality characteristics.In addition to this, there are few data on characteristics such as pressure and disturbance and it is difficult to handle, so if you are planning to use a numerical model, new facilities and capital investment may be necessary. Therefore, in is study, we classify the quality of the water and predict the quality of the water situation using Self-Organizing Maps.In this way, we can expect to help solution of the red tide outbreak mechanism, and expect to restrain damage of the red tide in predict the outbreak of the red tide beforehand.In the Self-Organizing Maps method, since new facilities and advanced mathematical knowledge are not required, it is possible to analyze the water quality situation from a comprehensive viewpoint simply by inputting the water quality data that has been acquired so far.

There are in great numbers of road structure built in highly economic growth period in Japan. Due to progressive deterioration of road structures, it is important to keep public facilities service level and ensure proper maintenance of overall facilities. In current maintenance work, road administration facilities are represented on a two-dimensional map, which is not suitable for pothole repair, inspection, or annual overhaul. In this research, the point cloud data of the road structures are constructed using terrestrial laser scanner (TLS) and the photogrammetry of camera on a Unmanned Aerial Vehicle (UAV). And, threedimensional maintenance of road structures was proposed using the combined point cloud of TLS and UAV. The proposal was conducted at Shiraito Highlandway in Karuizawa Town, Nagano Prefecture. The data were measured at 26 locations. The high density data can be acquired at about 10m radius from the setting spot of TLS, due to on-site environment and equipment performance limits.

Road supports our lives as an important infrastructure, and the responsibility for managing roads is crucial. The management of roads makes us a safe life by patrolling regularly to check the road deformation. However, regular patrols have been pointed out that it is extremely costly because people visit the work site. Meanwhile, with the development of laser measuring instruments, means for measuring three-dimensional shape on the ground as point cloud data are diversified at present. Therefore, these application to the management of roads is expected. And, it is considered that the deformation of the road can be recognized by extracting the difference from the point cloud data at multiple times. Therefore, in this research, we propose a method to evaluate differences of road object from point cloud data in multiple times.

In volleyball, it is important to grasp the movement of each player and analyze the formation. Therefore, the authors developed a system that tracks the movements of athletes using the game video. However, to acquire the tracking result for each specific player, it is necessary to recognize a unique uniform number for each player. In previous research, there is a technology to recognize printed characters and handwritten characters. However, it is difficult to recognize the uniform number when distortion occurs in the uniform number as the direction of the player's body changes during the game, or when the uniform number is not able to be visually recognized. In this research, we develop a system that distinguishes movements of specific players from uniform numbers and tracking results using a video of volleyball games.

In a previous study, we proposed a method of extracting road deterioration points using car probe data. In this paper, we collected car probe data and applied the proposed method with Fujisawa City as the target area. From the analysis results, road degradation points were extracted. Prior to detailed road deterioration diagnosis using MMS, we carried out a field survey by visual inspection. As a result, road deterioration was found at the extracted location. It was shown that the proposed method is useful, therefore, we promote refinement of the method.

Onomatopoeia is useful and intuitive expression for what is hard to describe sound, voice and state of things. Thus onomatopoeia is necessary in Japanese. Onomatopoeia is considered to have sound symbolism, which can make people estimate meaning or semantic usage of onomatopoeia. On the other hand, it isn't easy to estimate meaning and semantic usage of onomatopoeia for learners of Japanese language. In order to assist the learners to understand onomatopoeia, we have studied on auto-estimation of onomatopoeia's meaning and semantic usage so far. Whereas, the prior study dealt with the onomatopoeia regarding “human action”. This paper focused on the onomatopoeia of “natural sounds”, and discussed auto-classification on semantic usage of the onomatopoeia.

Onomatopoeias can simply describe sounds or state of things．Therefore they are used for conversation everyday in Japan．On the other hand，one of the problems that Japanese learners face is the lack of ways to learn onomatopoeia．This study proposes a thesaurus map which can describe semantic relationship among onomatopoeias．Our proposed method can transform the onomatopeia into a 2D-vector and assign it on the map．In our experiment，we examined whether the map can describe the semantic relationship as local distance on the map．The experiment utilized onomatopoeia samples which represent “human motion” to evaluate the map.

Onomatopoeia can simply describe sound or voice regarding nature, state and movement of things. Onomatopoeia provides rich emotional expression for Japanese books and everyday conversation. On the other hand such Japanese onomatopoeias are difficult for Japanese learners to learn. Besides that there are few ways to learn the onomatopoeias. This study proposes a thesaurus map which can describe semantic relationship among the onomatopoeias. Our proposed method can code the onomatopeia into a 2D-vector and assign it on the map．In our experiment，we examined whether the map can describe the semantic relationship as local distance on the map. The experiment utilized onomatopoeia samples which represent “sounds of things and tools” to evaluate the map.

In this paper, I propose a patent category estimation system using rough set theory. This system has a rule construction module and an estimation module. A rule construction module tries to decision rules from a decision matrix which is a document term matrix with category labels. It tries rules from lower approximate and upper approximate. An estimation module tries to estimate the category from unlabeled patent using decision rules. According to experiments with an expert, the maximum accuracy ratio is 62%. However, I cannot say this approach is better than the traditional method

We investigate regularization functions for rule ensembles, which have a form of weighted sum of rules. The rule ensembles are regarded as linear classifiers whose variables correspond to rules. Generally, there are tremendous possible rules in a data set. Hence, we should select a small subset of important rules. On the other hand, because of a large number of rules, some of rules correlate with other rules, i.e., the problem of multicollinearity arises. For rule selection, L1 regularization for a vector of rule weights is adequate, however, for multicollinearity, L2 regularization is required. In this study, to overcome these problems, we use a regularization function proposed by the author, which can adjust the size of rules in an ensemble.

The application of the obtained rules from table data sets to decision support is considered. Especially, the following three cases are focused on, (i) the character of the obtained rules, (ii) decision support in case that an obtained rule matches the condition, (iii) decision support in case that any obtained rule does not match the condition. We connect such cases with decision support, and realized an effective decision support environment in SQL.

Rule induction based on indiscernible classes from neighborhood rough sets is described in information tables with continuous values. An indiscernible range that a value has in an attribute is determined by a threshold on that attribute. The indiscernible class of every object is derived from using the indiscernible range. First, lower and upper approximations are described in complete information tables by using indiscernible classes. Rules are obtained from the approximations. A rule that an object supports, which is called a single rule, is short of applicability. To improve the applicability of rules, a series of single rules is put into one rule expressed in an interval value, which is called a combined rule. Second, these are addressed in incomplete information tables. Incomplete information is expressed in a set of values or an interval value. Two types of indiscernible classes; namely, certainly and possibly indiscernible ones, are obtained from an information table. The actual indiscernibility class is between the certainly and possibly indiscernible classes. The family of indiscernible classes of an object has a lattice structure. The minimal element is the certainly indiscernible class while the maximal one is the possibly indiscernible class. By using certainly and possibly indiscernible classes, we obtain four types of approximations: certain lower, certain upper, possible lower, and possible upper approximations. From these approximations we obtain four types of combined rules: certain and consistent, certain and inconsistent, possible and consistent, and possible and inconsistent ones. These combined rules have greater applicability than single rules that individual objects support.

In ordinary manufacturing lines, worker’s proficiency degree in the manufacturing operations was evaluated by a unit time or cycle time of making products. In this manner, it is hard to understand how the worker accomplishes the manufacturing operation and tasks with specific motions and objects derived from the hands, body and/or eyes. To overcome this limitation, we investigated the degree of worker’s proficiency of four elemental processes with methods of evaluation to estimate the connectivity between the sensory and motor integrations in human brain information processing. We measured eyes and body movements when workers manufactured products through three months. The developed method could differentiate the worker’s proficiency degrees between experts and novices. Coincidentally, expert workers had relatively high levels at every elemental process compared to novice workers who had low levels in two of the four elements. These results show the possibility that novices are not as proficient as experts when memorizing correct procedures as they are more likely to discriminate a specific point to accomplish tasks due to immature memory functions within the brain.

In order to clarify the cognitive skills of getting information by job hunting students, it is important to know what information they are focused on. The authors conducted experiments to measure eye movements while reading corporate data. As a result, when selecting a company, experienced people pay attention to continuing after entering the company and novice pay attention to whether treatment is good.

This paper deals with gaze behavior during walking in sightseeing areas. An experiment was carried out in which participants walk through a street with an eye tracker. It was revealed that residents tend to look into spaces in front of them whereas tourists tend to look at things around them.

We propose a method that evaluates the field from images without numerical information in RoboCup soccer simulation. Training data sets are generated by converting numerical field information into images and assigning corresponding target values according to the number of frames by the next goal. A convolution neural network is used for regression modeling where images are mapped into a scalar value. We discuss the way training data sets are generated and also the performance of the trained neural networks.

In the RoboCup Soccer 2D Simulation League, 22 software agents play soccer on a virtual soccer field. Each agent plays autonomously, making the 2D League a suitable test bed for studying multiagent systems. In this paper, we use an open program called agent2d. In agent2d, an agent selects an action using a search tree and an evaluation function that depends only on the ball’s position. We propose three modifications to this action selection. First, we added new terms that evaluate the state caused by an action and determine how precisely the action is executed despite interruptions by opposing players. Second, we prepared multiple sets of weight coefficients for the terms in the evaluation function. The sets of coefficients change depending on the ball’s position. Third, we developed a supervised learning system to ascertain the coefficients as the agents select actions directed by the human subjects. Our experimental results demonstrate the effectiveness of the proposed evaluation functions and learning system.

In the competitive team games such as soccer, the compatibility of tactics with opponent team greatly affects the result of game. In order to improve the team performance, teams need to adapt their tactics according to the opponent team’s one. We proposed the method to recognize opponent team using 3D-CNN that can deal with time-series data. In the experiment, we use the RoboCup Soccer Simulator as a environment. As a result, We could obtain about 80% recognition accuracy for 16 teams participating in RoboCup2017.

In such agent simulators whose action decision logic is defined by various interacting parameters, it is not clear, to what extent single parameter change affect on whole system behavior. For this kind of agent simulators, large scale micro simulation could be an effective approach to determine impact from parameter changes. In this paper, we verified the effectiveness of large scale micro simulations to evaluate an impact on whole system behavior caused from change of a specific parameter using RoboCup soccer simulator.

In this paper, we propose a method to extract two-tuples of specified frequency of sequence. Subsequences that appear frequently in sequence data are called frequent sub-sequences. This article focuses on the extraction of frequent sub-sequences from a given sequence. Since the occurrence rates of frequent sub-sequences are different from each other, it is desirable to extract only subsequences of a specific occurrence rate. To extract subsequences specified frequency, we propose a method using spiking neural network. This neural network can extract frequent subsequense of specified frequency by Spiketiming dependent synaptic plasticity learning rule and appropriate change of firing frequency of neuron. In simple experiments, the proposed neural network extracts frequent sub-sequence from test data including subsequences of various occurrence rates randomly. As a result, we confirmed that the proposed network can control the occurrence rates of extracted subsequences by adjusting the firing frequency of neurons.

In this article, we propose a new training method to train successive patterns for SpikeProp, which is a kind of spiking neural network. SpikeProp represents information by the timing of spikes and can train a transformation from an input spike sequence to the desired output spike sequence. In the previous study, the network tends to fail to output desired sequences in the case of successive input patterns with a narrow interval. We proposed a new training method that trains combined desired patterns and changes spike response function, which decides behavior of each unit. By simple experiments, we confirmed that the proposed method improves the network output for successive input patterns with a narrow interval.

Differential evolution is easy to implement, is a good performance optimization algorithm, and is applied in various ways. Candidate solutions in differential evolution are often initialized randomly, but search performance depends greatly on the initial Candidate solutions. It is possible to solve by introducing random elements such as mutation and noise in the evolutionary algorithm, but if introduced beyond necessity, the search speed will be lowered. In the proposed method in this study, we assume that an ideal search point group exists in the confidence interval, and randomly change the next search point candidate within the confidence interval. We confirmed that this proposed method improves the performance of the differential evolution in numerical experiments.

Denoising problems from time-series signals are important to get reliable data in the various fields (e.g., engineering, medical science and so on). The general methods for denoising are required knowledge about frequency bands causing noises. In contrast, a denoising autoencoder proposed in recent years is not required it. In this study, we propose a direct and parallel denoising autoencoder for high quality denoising algorithm. Here we describe proposed method and show its evaluation.

This paper deals with a design of the fuzzy model following servo system with disturbances. We construct the control system design using the easy algebraic algorithm of polynomial matrices on the differential operator. We guarantee the boundedness of the inner states for the control system. A numerical example shows that the output signal of the control system follows that of the model asymptotically in the case of the existence of disturbances.

In the super aging society of Japan in recent years, wheelchairs are cited as one of the welfare tools expected to increase demand. Power-assisted wheelchair has advantages that it can be used even by people with weak operating force and has the rehabilitation effect by operating push rim. However, since it can only be assisted with a constant assist ratio, it is likely to be jump-started or sudden stopped. In addition, its movement range is limited by the capacity of the battery.This study proposes a fuzzy inference type control method to realize the high power efficiency and high manpower efficiency driving, and verifies its effect by changing the operation pattern. The proposed control system generates the minimum assist torque necessary for acceleration to save power while the user is operating, and extends the coasting by adjusting the decrease amount of the assist torque after releasing the push rim. Furthermore, the power efficiency and manpower efficiency are evaluated with changing the operation pattern of the operating force and the time interval. As a result of the experimental verification in some high efficiency operating patterns, compared with the commercially available power-assisted wheelchair, it was shown that the improvement of the power efficiency and manpower efficiency can be realized.

This paper presents a path tracking stabilization control approach to unmanned aerial vehicles (UAVs) via rational polynomial fuzzy control. First, we provide a fuzzy polynomial model that is perfectly equivalent to UAV kinematics in wind environment. Next, we derive sum-of-squares (SOS) design conditions for the proposed rational polynomial fuzzy controller avoiding the real actuator saturation. The designed rational polynomial fuzzy controller realizes flight path tracking stabilization satisfying the real actuator constraint. Finally, simulation experiments are demonstrated to verify the effectiveness of the proposed design method.

A typical fusion of fuzzy inference and reinforcement learning uses a value-based method such as Q-learning assuming Markov decision process. On the other hand, we have proposed a fusion of fuzzy inference and policy gradient method, which is a policy-based and learns a policy directly, unlike a value-based method. The fusion uses a stochastic policy defined by Boltzmann distribution having an objective function consisting of the product-sum operation for membership functions and rule weights. Moreover, we proposed another objective function by using defuzzification based on a center of gravity model weighted stochastically and a constraint condition on the vibration of the output. In this study, we applied the fusion to simulations on speed control of an automobile and compared the objective functions. The results showed that the policies learned by our method, which uses center of gravity model and a constraint condition, tended to suppress vibration of the speed and accomplish the control task with a small number of steps.

Since the laser scanner used for continuous monitoring can acquire point cloud data with a high accuracy, it is expected to be utilized for construction management and grasping slope displacement. However, in order to realize these purposes, it is necessary to find a technique to detect the difference of ground surfaces in different time using point cloud data. As for existing researches, there are approaches for grasping secular changes of structures and visualizing soil volume in construction site. However, there has not any attempt to visualize difference of ground surfaces in every day. Therefore, in this research, we propose a method to detect difference of ground surfaces surveyed in different time.

ASCO-DAITO has introduced a handy laser measuring instrument since 2017. While the new technology application on site aimed to improve the work efficiency of topographical survey is examined, verifying self-location estimation in real time SLAM processing regarding the accuracy of this measurement instrument is also conducted. In this paper, the accuracy verification results by the handy laser measuring instrument and future application examples are stated.

It is necessary to construct three-dimensional data for realizing three-dimensional maintenance of the road. Three-dimensional data of road space are often constructed using Mobile Mapping System and terrestrial laser scanner. When acquiring three-dimensional point cloud data with a laser scanner, the laser hits the feature in front, data cannot be acquired for the features behind the sidewalk, etc. It is necessary to measure it several times. On the other hand, as one of the methods for constructing three-dimensional data, there is the SfM (Structure from Motion) technique. In this research, a method to generate three-dimensional data of road space based on SfM technique from video data taken with a commercially available video camera was studied for road maintenance.

Concrete placing is many manual operations such as operation of construction equipment, concrete placing depends on manual baced on empirical knowledge of skilled workers. For that reason, it is extremely important to inherit the experienced knowledge of skilled workers to young people. However, taking into consideration that a large number of skilled workers are expected to leave a lot, succession of skills is an urgent issue. In recent years, due to improvement of information and communication technology (ICT), we can measure detail of human behavior by utilizing this technology, we can measure, store and form empirical knowledge related to the human behavior of skilled workers. But, it is not clear that possibility of the measurement. The purpose of study is to make sure about possibility of measurement about empirical knowledge related to the behavior of skilled workers by using various measurement techniques．

In recent years, a system using Augmented Reality (AR) has been operated in various fields such as medical, architectural and entertainment. In general, a head mounted display (HMD: Head Mounted Display) used as an AR information presentation device is an optical see-through method, but since the background cannot be shielded, the visibility of the presented information depends on the brightness and color of the background. In this research, we propose a method to solve this problem by HMD control with special functions.

After-earthquake assessment of building in terms of safety and usability is performed by technicians who give their judgement based on in-field surveys and visual inspections. In this study, an efficient method is proposed for building damage detection and structural health monitoring system is introduced. First, the data of acceleration sensors that are placed each floor is processed for creating images which include shaking information of building during earthquake. Second, all images are divided into train set and test set. Train set is used to train a neural network by using deep learning and test set is used for evaluating model. Five different way of creating images and three neural network models are compared. Finally, quantity of data is extended by simulation and it is possible to determine the damage of the building in more detail. The result illustrates that two way of creating image and one model can be effectively used to detect building damage.

In large scale disasters, many evacuation sites should be set up immediately, and relief goods including foods should be being supplied to every site for a certain period.However, in 2016 Kumamoto earthquake, it is reported that relief supplies concentrated on a few large evacuation sites, and many of supplied foods became spoiled without consuming. However, at other sites, relief supplies were lacking. As this example, in large scale disaster, it is difficult to collect reliable information and to distribute necessary goods to appropriate site in allowable timing. To solve this problem, we employ the response threshold model for ant colonies, and we propose a sustainable relief goods distribution system in large scale disaster. To demonstrate the effectiveness of the proposed system, simulation was conducted.

This short note introduces essence of Uncertainty Theory and apply it to have more demand predictions of a new LRT transportation system designed by Utsunomiya city.

In this paper, we suggest a method of generating an emotional corpus customized according to the speaker. The emotional corpus is a database expressed with the word and the emotion values by fuzzy membership grade. The generation method is as follows. First, the system generates a basic emotional corpus by using the bag-of-words model. Then, by linearly transforming the basic emotional corpus, it generates an emotional corpus customized according to the speaker.

In this study, we propose the Kansei retrieval agent model with fuzzy reasoning and verify the optimization performance of the fuzzy rules. Kansei retrieval agent model is imitates the sensitivity of each user. Previous research has demonstrated the effectiveness in terms of presenting the user's preferences by optimizing only the membership function of fuzzy inference by numerical simulation and learning the user's evaluation criteria. However, we did not optimize the fuzzy rules of the model. Therefore, in this research, we verify the optimization performance of the Kansei retrieval agent model when optimizing the fuzzy rules using a numerical simulation. As a result, the proposed method shows to be effective from the viewpoint of user's evaluation criteria learning.

We apply acceptability functions, which is a new method for multi-objective optimization and multi-objective database retrieval, to a task of retrieving rental apartments and evaluate its effectiveness through human subjective tests.Acceptability function is a definition of human characteristics per each objective and defines how he/she accepts each objective along to its values. Total acceptability function is formed on a multi-objective space by synthesizing acceptability functions for all objectives and gives the total acceptability value to each of retrieved apartment, and we can use them to the display ranking of retrieved apartments.Two experimental evaluations are conducted. The first one is that human subjects compare the proposed apartment retrieval system that gives ranks to all retrieved apartment and a conventional retrieval system that outputs all apartments satisfying search conditions without ranks. The results has shown that the proposed one was more preferable. The second evaluation is to check how the ranking from our system and that from human subjects is similar. Spears' ranked-correlations has shown their high similarity.

The demand for wheelchairs is predicted to increase due to lack of caregivers and aging society. Electric power-assisted wheelchair is one of the support equipment which can easily move long distance with the small pushing force. However, since a commercially sold power-assisted wheelchair is designed for everyone, if a person who prefers the fast moving speed uses it, he/she needs to adjust the pushing torque to produce the preferred speed. Thus the comfortable driving with a commercially available power-assisted wheelchair depends on the user's operation skill and the physical and mental burden is not small. Therefore, this study designs a preference estimation map in advance through the preliminary driving and SD method and set as some appropriate target values of the driving control system. The driving control system for determining the assisted torque is constructed by fuzzy inference to realize the driving target values based on the preference estimation map. Some driving experiments to compare the commercially available wheelchair and the proposed method are performed to verify the effectiveness of the proposed control system.

We aim to develop a swimming motion coaching system for the beginners and intermediate swimmers by using one inertial sensor. This system requires the process of automatically estimating and dividing the section of swimming motions from the data. In this paper, as the first step, we constructed a classifier of swimming motions of the butterfly based on the random forest and estimated its turn section. As the result of verification with the test data, it was possible to estimate the start and end points of the turn and their errors are .322±.251s and .416±.186s, respectively. These results suggest that our method can be used in the swimming motion coaching system.

Aiming the extraction of the significant solar wind parameters which trigger the geomagnetic disturbances (disturbances of AL-index), we try to test an extraction of the most influential parameter of solar wind which would cause a magnetic storm seen on 8th September 2017 based on a neural network. Terrestrial magnetosphere is always exposed and disturbed by high-speed plasma flows from the Sun (solar wind). Large-scale geomagnetic disturbance bring various troubles to the electronic instruments in ground. It is important to specify what the solar wind parameters trigger the significant geomagnetic disturbances. To extract the solar wind parameter(s) to trigger a significant geomagnetic disturbance, we adopted a neural network called “potential learning (PL)”, which can predict targets and interpret a model. Utilizing the solar wind parameters, we created a prediction model whose generalization performance, evaluated by correlation coefficient between targets and predictions, was 0.8195. As a result, we can extract a parameter of “dynamic pressure”, obtained by solar wind velocity and number density. Because, in this case, it can be considered that the magnetospheric squeezing would play a role in triggering the AL-index disturbances, this extraction of “solar wind dynamic pressure” due to PL should be a reasonable result. Therefore, we conclude that the PL is useful and important tool in extracting what solar wind parameters affect the significant geomagnetic disturbances.

Deep learning technology becomes popular to realize automatic polyp detection in the medical diagnosis. This paper proposes a method using transfer learning applied for the endoscope image by introducing fine turning with CNN which gives the high evaluation in the general object recognition. Proposed method uses DenseNet , AlexNet and VGG16 models and evaluates the accuracy to each model which is learned with ImageNet, and constructed effective classifiers for polyp detection. Experimental result suggests that the proposed method improves the classification accuracy using CNN feature and SVM.

Non-safety behavior in car driving such as smartphone operation, mobile phone call, etc. are major concerns that lead to heavy traffic incidents where the number of such incidents is increasing recently. This research proposes a method that jointly estimates head posture as observation action and hands behavior as one of the operation actions in a form of joint probability distribution based on state space modeling formulation. Source of information consists of two videos captured by facial camera and steering operation camera, where the facial camera is set over the steering column to capture frontal face image of a car driver and the steering operation camera is set above seatbelt roller to capture steering area. Furthermore, utility of the estimation result to detect non-safety behavior in car driving such as smartphone operation has been considered.

This research is intended to investigate the scene that user feels fun as a part of study to make a game fun. We made OZ-RP system to make Robocup Soccer Simulation 2D video game for experiment. In this paper, we report about the result of the subject of an experiment that aim to investigate fun scenes.

This study aim to find out the reason of the strength of M-UCT that UCT with the node using the moves in Turn-Based strategy games. We propose S-UCT that M-UCT with unit abstraction. The proposed method played against M-UCT, and the winning percentage was about 30%.