To prevent accidents caused by human factors, it is essential to consistently monitor driver’s state and alert them to potential hazards. The authors have been developing a wrist-worn sensor-based method of driver status monitoring. Our previous research has used the data collected using a driving simulator in a laboratory environment. However, because the real-world conditions include vehicle vibration and diversity of driver behavior, etc., collecting the data under the actual driving environment is essential. Therefore, we have developed and evaluated an in-vehicle measurement system to collect data to estimate the driver’s state in actual driving conditions. Our system allows the user to easily collect up to 120 minutes of synchronized multi-modal data, i.e., hand acceleration, vehicle acceleration, position, steering angle, and video. In this short note, we report on the evaluation result of synchronization accuracy and precision of steering angle estimation based on the data collected using our developed system.

We propose an illustration editing system using onomatopoeia related to color features for beginners of image editing. The proposed system employs two phases: tuning of amounts of color features change for each user sense of various onomatopoeias and editing image with the tuning results. The tuning phase uses neighboring search method to tune relations between user sense and each onomatopoeia imagine. We conducted evaluation experiments for investigating the effectiveness of reducing user image editing loads. The results showed that the proposed system was more effective for reducing user editing loads than the conventional system (editing image by changing each feature manually).

This short note discusses a design of a robust stabilization controller for Takagi-Sugeno fuzzy system with uncertainties in control matrices. Takagi-Sugeno fuzzy system can represent a wide range of systems, and due to this, it is effective for nonlinear system control. However, generally on system identification, it is impossible to make math equation model which is completely same as control object. Therefore, we need to consider designing a controller for Takagi-Sugeno fuzzy system having uncertainties as modeling errors in order to improve control quality of the controller. In this short note, we employ a non-quadratic Lyapunov function including integral functions of membership functions. Finally, a numerical example is given to illustrate the effectiveness of our designed controller.

The phenomenon in which a sound itself evokes a certain image is called sound symbolism. Many studies have been reported regarding sound symbolism, but there are many remained aspects of sound symbolism that have yet to be clarified. The final goal of our study is to clarify the whole picture of sound symbolism. To reach this goal, we collect a wide variety of examples. We will then attempt to discover their unique sound symbolism. By integrating the many examples of sound symbolism obtained, we attempt to clarify the whole picture of sound symbolism. This study adopts monster names in the Monster Hunter series, which are considered to be examples of sound symbolism. We consider that sound symbolism might exist between the monster name and the total length of the monster. In the experiment, the classifier was constructed on a CNN basis and Grad-CAM, an XAI technique, was used to analyse the acoustic characteristics of the monster names. We hypothesised from the experimental results that the /r/ sound gives the impression of being louder. We also tested the hypothesis in a subjective evaluation experiment and found that the results statistically supported the hypothesis.

The Internet is becoming more common as a means of collecting tourist information. Traditionally, travel information has been collected from travel magazines, television, travel agencies, etc. However, due to the widespread use of the Internet, tourist information is being acquired by searching the Internet instead. In addition, people who visited tourist spots wrote their experiences on his/her SNS. Information analysis on the Internet is becoming necessary to increase the number of tourists. In this research, we collect regional information from the huge amount of data that exists on the Web for use in tourism promotion and regional promotion. We perform text mining analysis on the collected data. The purpose is to extract useful knowledge and knowledge based on the results. We analyze the differences in tourist information between the corona disaster and the non-corona disaster. We examine and compare whether there are differences in the information disseminated by SNS and other means.

This study aims to establish a method with ordinary videos (without special equipment) for gait quality assessment. The treatment for Cerebral Palsy, which is a movement disorder, requires gait quality assessment routinely. However, the current assessment methods need expensive equipment and high technical knowledge of rehabilitation. This paper aims to develop a system to evaluate a patient’s gait without special equipment. We propose a method to estimate the gait quality using off-the-shelf human pose estimation and Auto-Encoder. Evaluation experiments using actual patients’ data were conducted to discuss the effectiveness of the proposed method. The correlation coefficient between the proposed method and the typical gait pathological index suggests that the proposed method has enough capability to estimate the patient’s gait abnormality.

In this study, we explored the influence of robot’s gaze on human behavior in interactions when passing each other. Robot gaze affects the distance between subjects and the robot. Therefore, we analyze the distance that the subject takes toward the robot depending on whether the robot’s gaze is present or absent in the passing-by environment, as well as the subject’s gaze data. Using the analysis results, we aimed to evaluate whether the robot’s gaze affects the distance to a human in a passing environment. We devise a design of the robot gaze at people when passing each other. From the experimental results, it was confirmed that the robot does not affect the distance between subject in the passing environment. However, the robot’s gaze attracts more attention from the subjects. Therefore, the design of the robot that gazes at the human at 5.5 meters around the robot may trigger an interaction between the human and the robot when passing each other.

In order to reduce the risk of children becoming mentally ill, we develop a system to estimate the user’s psychological state based on the tree test, which is one of the projective psychological tests, and provide psychological education that matches the result. In this paper, we developed the module to estimate the user’s psychological state based on the tree test. Users can choose from 2D objects such as tree trunks, roots, and leaves to complete the drawing of the tree. Personality and psychological tendency are estimated from the result of the drawn tree. An experiment was conducted to compare the tree drawn by the proposed system with the one drawn with a pencil on paper. As a result, while the user was able to draw the tree that the user expected using the system, there was a difference between the drawing of the tree on paper and the one drawn by the system.

Various multi-label classifiers have been proposed for multi-label classification problems. Our previous study has proposed an adaptive resonance theory (ART)-based clustering method using correntropy-induced metric (CIM) as a similarity measure, called CIM-based ART for Multi-Label Mixed Data (CA-MLMD). CA-MLMD adaptively and continually generates nodes corresponding to input data, and the generated nodes are used as a classifier. Moreover, CA-MLMD learns new data and label information continually and handles mixed datasets that contain both numerical and categorical attributes. However, CA-MLMD is highly affected by local data points around the node in learning categorical attributes, which may deteriorate classification performance. This study proposes CA-MLMD-weight (CA-MLMD-w), which uses weights defined by categorical attributes of each node and reduces effects of local data points by considering categorical attributes of the entire data. Numerical experiments on real-world datasets show the effectiveness of the proposed method.

The purpose of this study is to investigate the effect of improving safety confirmation behavior by instruction including compliment by a driving support agent. We conducted an experiment using a driving simulator and an eye tracking device to investigate changes in the safety checking behavior of drivers before and after experiencing a driving support agent that gives compliment and guidance. The results of the experiment showed that the number of situations in which drivers improved their safety-checking behavior with compliment was greater than that without compliment, suggesting the possibility of improving safety-checking behavior.

The “Noto Satoyama Satoumi Go” is a sightseeing train operated by the Noto Railway. This train operates on Saturdays, Sundays, and public holidays through the world agricultural heritage site of “Noto Satoyama Satoumi” in the Noto Peninsula. Although this service is availed by a considerable number of group tourists in partnership with travel agencies, the number of individual tourists on this train has been decreasing annually, thereby posing a challenge to disseminate information to individual tourists. Promotions conducted using pamphlets and videos in the past have not been sufficiently effective. Therefore, more presence and memorable means of disseminating information should be realized. To address this requirement, we developed a virtual reality system to boost tourism by simulating the train-ride experience. The results of this study imply that the information provided by the system will be remembered by the users. Furthermore, this system increased users’ intention to use the sightseeing trains.

Anomalies in industrial images can be categorized into logical anomalies and structural anomalies. Logical anomalies refer to irregularities such as object deficiency, excess, or misplacement, while structural anomalies indicate impurities such as dirt, scratches, or foreign matter inclusion. In conventional anomaly detection methods based on normalized flow, variable transformation is performed considering local information in the feature map. While these methods generally have good detection performance with respect to structural anomalies, they are not good at detecting logical anomalies. In this study, to address this issue, we propose the Multi-head Self Attention Flow (MSAFlow), which introduces a self-attention mechanism into the normalization flow to capture the relationships between features during variable transformation. The proposed method was evaluated in comparison with the conventional convolutional layer-based normalization flow on the MVTecLOCO dataset, achieving a 5% improvement in the average AUROC across all categories.

Multi-objective fuzzy genetics-based machine learning can efficiently obtain a set of fuzzy classifiers considering the maximization of classification performance and the minimization of the model complexity by using an evolutionary multi-objective optimization method. However, multi-objective fuzzy genetics-based machine learning has a strong bias towards minimizing complexity in the optimization process, making it difficult to generate classifiers with high classification performance. In our previous study, two-stage fuzzy genetics-based machine learning has been proposed to mitigate this bias: first, an accuracy-oriented single-objective optimization is performed, and then a multi-objective optimization is performed to maximize the classification performance and minimize the complexity. The use of an archive population has also been proposed to obtain a better set of classifiers in two-stage fuzzy genetics-based machine learning. However, the effects of the use of an archive population on a set of classifiers obtained by two-stage fuzzy genetics-based machine learning have not been fully investigated. In this paper, we investigate the effects through computational experiments on a wide variety of real-world datasets.

Reinforcement learning methods include learning a simple and accurate dynamics model of the environment (world model) and performing trial-and-error in a compact latent space. However, because the world model is learned using reconstruction errors, performance deteriorates when the visual environment becomes more complex. In contrast, learning the world model by contrast learning reduces the performance degradation even when the visual environment is complex. However, there is still a problem of performance degradation when the batch size is reduced. In this study, we propose a method for learning world models using non-contrast learning. We believe that this method can solve the problem of performance degradation in tasks with complex visual environments. In order to improve robustness with respect to visual information, we introduced a loss function that suppresses the effect of background information that is irrelevant to the task. As a result, the proposed method performed better in 4 out of 6 tasks with a normal background, and in 5 out of 6 tasks with a complex background.

This study investigates word segmentation method based on NPYLM for minority language texts. Applying conventional segmentation methods to text in minority languages is difficult due to the lack of knowledge for the language and insufficient quantity of the text. In particular, NPYLM, one of the conventional methods, can segment texts without prior knowledge, but tends to cause over-segmentation in the case of insufficient data. In this paper, we propose a two-step NPYLM to improve the over-segmentation. First, the first NPYLM is trained a given text with NPYLM to obtain replacement candidates. Next, each candidate words is replaced by different single character. Then, the second NPYLM is trained replaced text. Finally, we get a segmentation result with less over-segmentation. Experimental results show that the proposed method improves the F-measure (based on segmentation position) and the average word length for texts in English, Japanese, and a minority language. Experimental results show that the proposed method improves the over-segmentation for texts in English, Japanese, and a minority language. We conclude that the proposed method brings effective performance for various languages.

In order efficiently to explore and exploit the environment humans and animals (i) generalize experience through trial-and-error in similar states and (ii) focus on a certain level of achievement to switch risk attitudes in an appropriate way. In that process they build up and draw on a model of the environment that enable them to deal with contextual information such as the features of each option and the current state of the environment. In this paper, we propose the Regional Linear Risk-sensitive Satisficing (RegLinRS) model, which is a decision-making algorithm that incorporates the exploratory strategy and risk attitudes of humans and animals. Based on the simulations of the contextual bandit problems, we evaluate the performance and exploratory efficiency of RegLinRS. We then analyze RegLinRS’s adaptive rationality, that is, in what manner and to what degree the algorithm works to promote the survival of agents. The results show that RegLinRS enables the agent to quickly secure a satisfactory goal while minimizing expected loss.