The purpose of this study is to investigate the effect on the improvement of user's safety confirmation behavior by checking whether the user is confirming safety confirmation while driving, and by providing guidance including praise according to the degree of safety confirmation. We developed a system in which a driving guidance agent judges the user's safety confirmation behavior using face orientation estimation, and then gives guidance, including praise, according to the degree of safety confirmation and the driving situation. We conducted an experiment to investigate the change in safety confirmation behavior before and after the user experiences the driving guidance agent using an eye tracking device, and whether the safety confirmation behavior is sustained, and compared the effects of praise and no praise guidance, respectively. The results of the experiment showed that there was a possibility that the safety-checking behavior was improved and sustained better with the praise instruction than with the no-praise instruction.

現在「AI」と呼ばれる深層学習は数億のパラメータと数百億のトレーニングデータによって高い実用性を持つモデルの学習を行い注目を浴びている。いっぽうファジィシステムも高い表現力を持つが可読性が特徴であるため、かえって大規模データ、大規模ルールのシステムを扱うことは忌避されてきた。本発表では引くくらい大規模なルール数のファジィ推論の計算の検討について報告する。

The phenomenon in which a sound itself evokes a certain image is called sound symbolism. There are many aspects of sound symbolism that have yet to be clarified. The aim of our study is to clarify the whole picture of sound symbolism. For this purpose, we collect a wide variety of examples of sound symbolism and attempt to discover their acoustic features. In this study, we employed the names of monsters in the Monster Hunter series as examples of sound symbolism, and investigated the acoustic features of the differences in the names of monsters according to their body size. Machine learning classification is used to investigate and verify the existence of sound symbolism.

The phenomenon whereby a sound itself gives a particular impression is called sound symbolism. Although sound symbolism has been reported in various fields, the full picture has not yet been elucidated. We are therefore collecting and investigating examples of sound symbolism in order to clarify the overall picture of sound symbolism. In this study, we have taken up the names of Takarazuka Revue members as examples. We believe that the naming of Takarazuka Revue members may have a unique pattern. We investigated their acoustic features by using machine learnikng. In our study, a Convolution Neural Network (CNN) was used for the investigation. Based on the investigation result, we analyzed the acoustic characteristics about the names of Takarazuka Revue members.

In recent years, research and development of robots that can coexist with humans in various environments has been conducted. Among these, various psychological experiments are being conducted to investigate the psychological state of people in response to the robot's speech and actions. For example, it has been reported that people's motivation increases when they are praised by a robot, and that people who are praised become more complimentary toward others. On the other hand, there are few examples of analysis of human behavior in robot situations. This study examines how university students behave when they fail in a task and find themselves in a difficult situation. The experiments conducted an observation experiment in which a university student helped or comforted a robot. The situation of experiment was the robot had to answer all five questions correctly. We analyzed the behavior of each subject when the robot cannot correctly answer the last question.

A speech recognition mask using conductive knitted fabrics enables voice input by mouth movements (silent speech). The conductive knitted fabric sewn into the mask acts as a flexible and highly sensitive strain sensor to measure the deformation of the mask caused by mouth movements, and the data obtained is used to estimate the silent voice. Using this mask, users can input voice without worrying about ambient noise or privacy issues. We conducted some experiments of voice recognition using the developed mask to demonstrate its effectiveness.

In this research, we have successfully developed an advanced system for recognizing surroundings, designed to support visually impaired individuals in navigating through uncrowded areas, effectively reducing the risk of their white canes colliding with other obstacles. Our primary objective was to overcome the limitations of previous research by creating a portable system that eliminates the requirement for additional hardware purchases. To accurately measure distances to individuals in the vicinity, we constructed a system utilizing iPhone LiDAR technology. Additionally, we incorporated YOLO, an object detection algorithm, to effectively recognize people in the surrounding area. In an evaluation experiment, a visually impaired individual utilized our system to measure both the time taken to reach their destination and the frequency of contact with other individuals while in motion. The results suggest that when using our system in comparison to rely solely on a white cane, the user’s walking speed improved, and they were able to navigate without making physical contact with others. By conducting these experiments, our aim is to enable individuals to walk safely by providing them with positional information while ensuring a comfortable experience for both the user and those around them, avoiding any instances of discomfort or collisions.

Recently, research on robot communication has been conducted. Robots have been shown friendly on the physicality and developed positive relationships through human contact. On the other hand, virtual agents(VA) have an advantage that is portability and taking a communication complementary with robots. In our research, we aim to develop a robot-agent integration system that enables continuous communication between robot and VA. By sharing conversation content and user parameters, the system allows for ongoing communicative responses between robot and VA. Furthermore, we perform experiment to determine whether the developed system meets the requirements specifications and conducts customized communication for the user.

In this study, we propose a system that optimizes the gestures used to express the personality traits of a communication robot according to user preferences. In order for a communication robot to make people feel comfortable, it is necessary to convey a friendly impression. However, conventional robot personalities are designed based on general human impressions and do not take individual users’ preferences into account. In order to tackle this challenge, we present a system that utilizes Interactive Evolutionary Computation (IEC) to optimize the gestures employed to convey the personality traits of the robot, aligning them with the user’s specific preferences. The effectiveness of the system was then verified through an experimental evaluation. As a result, it was suggested that the proposed method could help the robot convey better impression to users, thereby improving its familiarity.

A previous study has reported that displaying a static target point (fixation point) at the center of the virtual reality screen can reduce virtual reality sickness. This paper investigates the effects of moving the target point. We conducted an experiment comparing three conditions: (1) the following condition where the target point trailed the user’s gaze, (2) the preceding condition where the target point moved ahead of the gaze, and (3) the control condition with the fixation point. The results showed that the following condition increased nausea and fatigue compared to the control condition. On the other hand, no significant differences were observed between the preceding and control conditions. These results suggest that using the preceding target point may reduce virtual reality sickness as much as using the fixation point.

This research aims at the analysis of effects on work efficiency during listening to stress or relaxation sounds. The work efficiency in this study is defined as the speed and accuracy of work. In order to measure the work efficiency, we evaluated the effects of listening to stress and relaxation sounds using the Kraepelin test, which is often used in the field such as employment and qualification examination. Generally available music was used for the sound to be heard by the subject, and heart rate information and sweating information were the measured as an index of relaxation and stress during work. As a result of the experiment, the effects of stress and relaxation sound on the human work efficiency were analyzed, and it was confirmed that there was the strong correlation.

Local governments are actively engaged in regional promotion activities, and the creation and utilization of local government PR logo marks is also becoming active. Local government PR logos are printed on posters and wrapping paper for souvenirs, and are used in goods that use the logo mark. In this study, we conduct an impression survey on local government PR logos, and clarify the impression space and evaluation axis in local government PR logos. Next, we conduct a survey on the combination of logos and goods to clarify the elements of logos that are suitable for apparel goods.

This study proposes a fuzzy clustering algorithm for vectotial data, which is constructed by extending the fuzzification parameters in the q-divergence-based fuzzy c-means algorithm (QFCM). The proposed algorithm, extended QFCM (eQFCM), is an extension QFCM and the penalized fuzzy c-means algorithm proposed by Yang, referred to as Y-type FCM (FCM). eQFCM extends both the two-parameter QFCM and Y-FCM algorithms to a four-parameter model. Through numerical experiments using an artificial dataset, the theoretical discussion is substantiated and some numerical experiment using real datasets show that the proposed algorithm outperforms conventional ones in terms of clustering accuracy.

Fuzzy clustering algorithms that extend mixture models of Gaussian and t-distributions have been proposed under various fuzzification. On the other hand, multivariate power exponential distribution has been applied only to probabilistic mixture models and has not been extended to fuzzy clustering. In this report, we propose five fuzzy clustering algorithms that extend the mixture model based on multivariate power exponential distribution by following KL-divergence regularization type, Fuzzy Classification Maximum Likelihood, Tsallis-entropy regularization type, q-divergence regularization type, and Bezdek type. The effectiveness of the proposed algorithms is verified by numerical experiments.

Various fuzzification techniques have been applied to clustering algorithms for vectorial data, such as Yang-type fuzzification and extended q-divergence-regularization, whereas only a few such techniques have been applied to fuzzy clustering algorithms for series data. In this regard, this study presents four fuzzy clustering algorithms for series data. The first two algorithms are obtained by penalizing each optimization problem in the two conventional algorithms: Bezdek-type fuzzy dynamic-time-warping (DTW) c-means and Bezdek-type fuzzy c-shape, with the cluster-size controller fixed. The other two algorithms are obtained from a conventional algorithm, q-divergence-based fuzzy DTW c-means or q-divergence-based fuzzy c-shape, by distinguishing two fuzzificators for membership from those for cluster-size controllers. Numerical experiments are conducted to evaluate the performance of the proposed algorithms.

Fuzzy c-means is a basic and general fuzzy clustering algorithm for vectorial data and several variants of this algorithm have been proposed. However, research and development of fuzzy clustering for series data is not as progressive as that for vectorial data. In particular, research on fuzzy clustering algorithms using series models is not advanced. In this work, we proposed several fuzzy clustering algorithms based on a combination of four types of series models, namely the autoregressive and moving average model, generalized autoregressive conditional heteroskedasticity model, hidden markov model, and linear Gaussian state space model, as well as three types of fuzzification techniques, namely Kullback-Leibler divergence regularization, Bezdek-type fuzzification, and q-divergence basis.

The Kestenberg Movement Profile (KMP) is a method of movement analysis from a psychoanalytic perspective. In KMP, the analyst usually makes a recorded waveform (notation) called a rhythm line from the subject's natural daily movements, and the basic pattern (pure rhythm) extracted from the rhythm line and other characteristics are used for analysis. However, the obtained rhythm lines contain ambiguities because the rhythm lines are drawn based on the analyst's physical empathy, senses, and habits. Extracting and analyzing pure rhythms requires a wealth of experience and a high level of expertise. Therefore, the demand for a system to automatically extract and analyze pure rhythms from the given rhythm lines has been growing in the field of clinical application. In this study, the authors first developed a pen tablet-based system to input rhythm lines toward the automatic movement analysis using KMP. As a first step, we investigated a feature extraction method for extracting pure rhythms from rhythm lines drawn by an analyst.

In polysomnography, a sleep technologist visually analyzes a time series of respiratory movements, i.e., the respiratory curve, measured with a RIP belt or other equipment to record abnormal respiratory events such as apneas and hypopneas. The present study aims to develop a method for detecting peculiar patterns in the respiratory curve to automate this analysis process. Our approach here is to divide the respiratory curve into partial time series using Singular Spectrum Transformation, then classify them based on the similarity of the time series patterns to identify essential features that characterize abnormal respiratory motions. This paper presents our investigation into applying the respiratory motion patterns obtained by a time series clustering method to detect abnormal respirations.

Non-negative matrix factorization (NMF) is a basic method for analyzing the intrinsic structure of such non-negative matrices as environmental observation data, but cannot work well when datasets include incomplete data subsets drawn from different generative schemes. This paper proposes a novel switching NMF algorithm with partial distance strategy, which simultaneously achieves construction of cluster-wise submodels and handling of missing values.

Parametrial invasion (PMI) is a key factor to propose a treatment plan for cerrvical cancers. This study proposes a machine learning based PMI evaluation method to determine the presence of PMI in cervical cancer independent of the physician’s diagnostic experience. The proposed method requires the T2-weighted MR images, its cancer tumor mask, and dilated mask to compute radiomics features. The Lasso regression extracts important features, and they are used to train machine learning models and to classify PMI presence. Experiments on 10 PMI patients and 35 non-PMI patients showed that the logistic regression achieved the highest classification performance among the tested 8 classifiers. The comparison with physician’s assessments, the proposed method obtained superior results.

Ureteral stone is a disease in which crystals form in the urine and form stones in the ureter. In recent years, extracorporeal shock wave lithotripsy (ESWL) and transurethral lithotripsy (TUL) have been established as the principal treatment methods for ureteral stone. ESWL is less physically burden, but its success rate is lower than that of TUL. Physicians select an appropriate treatment based on clinical findings, CT images and so on. However, the success rate of ESWL is still about 70%. The purpose of this study is to reduce the number of patients who suffer the physical and financial burden of double treatment due to ESWL failure through the clinically interpretable and highly accurate predictive model of ESWL outcomes. We collected X-ray and CT images and clinical findings from 162 urolithiasis patients. First, the proposed method extracts image features using deep learning and hand-crafted shape and texture features. Next, it constructs a model to predict ESWL outcomes using machine learning. The proposed model performed with an accuracy of 0.887, a specificity of 0.620, an AUC of 0.890 by using a logistic regression model. We then analyzed the influence of each feature on prediction by using SHAP values. Also, features are selected by using SHAP values to improve the performance. We achieved an accuracy of 0.913, a specificity of 0,857, and an AUC of 0.956 by using random forest where seven features were selected. The experimental results indicated that we were able to reduce the number of features and develop a more interpretable model.

Total knee arthroplasty (TKA) is a surgery to decrease knee pain and to improve walking ability by replacing a knee deformed by osteoarthritis or rheumatoid arthritis with an artificial joint. The patient satisfaction of TKA reported to be 75% to 89% and was lower than total hip arthroplasty (THA). The patient satisfaction can be quantified by using postoperative patient satisfaction called KSS2011. Conventional studies have investigated patient satisfaction prediction using preoperative patient information. In addition to the preoperative data, this study employs two kinds of intraoperative data after removing anterior crucial ligament (ACL), and after implanting prothesis according to the progress of TKA surgery. It enables us to optimize intraoperative conditions to maximize the patient satisfaction during TKA operation. We introduce two kinds of feature extraction methods, and three kinds of models to predict the patient satisfaction after TKA. The experimental results on 62 patients (male:17, female:45) showed that KSS2011 prediction with the proposed method achieved the minimum root-mean-squared-error (RMSE) of 6.15, 6.05. 5.75, on preoperative, after ACL removal, and after implanting, respectively. Furthermore, we investigated the importance of features with SHAP. It indicated that implant GAP was the most important feature to predict patient satisfaction.

Lifestyle Diseases are caused and progress due to lifestyle behaviors encompassing dietary patterns, physical activity, and cigarette usage. Lifestyle diseases are characterized by their insidious nature, often lack discernible subjective symptoms, and increase the likelihood of severe illnesses going unnoticed. Consequently, preventive health care, entailing health check-ups and comprehensive medical examinations, garners escalating consideration for the timely detection, treatment, and prevention of severe illnesses. In this paper, we propose a Transformer-based method for predicting diabetes, which is a typical lifestyle disease, to predicate on medical data obtained from various health check-ups and other sources. Furthermore, we evaluated our method that appropriately accounts for the missing values inherent to medical data, a distinction stemming from differences in the number of examination items and the frequency of medical examinations. The Transformer-based method shows 83.2% of the AUC score.

To resume a work after it has been passed several days or weeks, much thinking about the work is often forgotten. “Thinking about the work” can be replaced a ‘context,’ that is the background or situation regarding the work. It is important to reconstruct the ‘context’ when we come back to the suspended task. The system proposed in this study focuses first on soliloquies that are uttered while working alone. Most of the soliloquies during the work act to organize or confirm what they are thinking. It is expected that the record of soliloquies will play a role as a note of the work. The recorded audio is presented in the form of text boxes. We will facilitate viewing, such as highlighting keywords according to the importance of the word. Second, we consider acquiring screenshots as history of the operation of the PC to record operation and viewing. We focused on the operation history because the operated part could be a record of user’s thinking. In addition, we will also consider user manipulation of text boxes, such as reordering. We conducted the preliminary experiment for the next experiment. We tested two kinds of the tasks; preparing a material to introduce an academic English paper and coding as the working task. We recorded voices and screen shots simultaneously. We classified the voices converted in texts and examined the result to prepare for the following experiments.

The atmospheric density in very low Earth orbit (VLEO) is a crucial factor for nano/small satellite missions. However, the profile of the atmospheric density is not clear yet because of the difficulty to obtain data in VLEO. In the duration of 2016—2017, The deployable nanosatellite EGG mission was conducted, and its sparse coordinate data was obtained by the global positioning system (GPS). To utilize this data for atmospheric density estimation, we constructed a methodology to estimate atmospheric density based on Gaussian process regression and Bayesian optimization, and estimated atmospheric density at altitudes around 200 km.

The inner-class model is introduced as a basic principle of artificial intelligence. More flexible, human-like, and context-based performance is expected for this model, such as with respect to recognition, control, and problem solving. The only assumption of this model is that a feature corresponding to the co-occurrence of two features is generated by just pairing the two features. An arbitrary feature created by this assumption is called an agent. This paper describes how the model 1) records and reproduces time-series data, 2) sets and reaches goals, 3) creates new concepts, and 4) deletes useless agents. The results of experiments are also shown to verify that the model is able to record and reproduce a simple time-series of data, and that the model reaches a given goal.

In this paper, we consider large matrix which is almost dominated by randomness, but also contains small amount of important information. For the improvement of approximation of eigenvalue distribution of Gram matrix of Bi-correlated model, the aim of this paper is to propose evaluation methods for the approximation. This will be used for the next step of the model improvement of the matrices and its application to detailed analysis of principal component analysis.

The purpose of this paper is to understand the relationship between the physical parameters which show the solar wind conditions (e.g., magnetic field, solar wind velocity, and plasma number density) and global geomagnetic conditions represented by Kp (2− ∼ 5+). The solar wind parameters observed by satellites over 22 years from 1998 to 2019 were used, and they were analyzed by neural networks that focus on the potentiality of input neurons (potential learning; PL). As a result, we found that the solar wind velocity and the plasma number density may play an important role in the Kp index whose range is from 2− to 5+.

Visually impaired people face a variety of problems when traveling outdoors. For this reason, research on systems to assist the visually impaired in walking safely has been conducted in recent years. Most of them focus on obstacle avoidance and use existing navigation services as they are for guidance. However, such systems cannot provide guidance to pinpoints such as building entrances. Therefore, we propose a method of guiding the visually impaired to building entrances using a recurrent model. Our system uses a recurrent machine learning model to output optimal voice instructions based on the user's walking trajectory and time-series data of entrance locations detected by YOLO. By using time-series data, the system can accurately correct the direction of the entrance even if the entrance is temporarily out of the camera's angle of view. Future work includes consideration of the walkable area and implementation of the system in a server-client format.

Spiking neural networks (SNNs) are neural networks (NNs) that use spikes (pulses) as input and output signals. Since their units have the ability to handle time-series information, SNNs are expected to be applied to complex time-series signal processing. However, the complexity of the unit operation makes it difficult to scale up the network compared to non-spiking NNs. Therefore, we pay attention to the fact that simplified activation functions (such as ReLU) improve the performance of large scale NNs. We discuss the effect of simplifying the spike response function, which corresponds to the activation function in SNN, from the viewpoint of learning performance, and report the results.

Neural networks have been a staple of the machine learning community since their inception decades ago. They have enjoyed a further surge in popularity as hardware and data limitations have allowed for the creation of deep neural networks, which have shown remarkable results in fields such as reinforcement learning and image classification, among others. However, the black box nature of neural networks presents a hurdle to their adoption, particularly in safety-critical domains such as medical diagnosis. A natural approach to explain the behavior of neural networks is the extraction of an interpretable set of rules which sufficiently mimic the behavior of the neural network. In this review, we provide an overview of the latest research in the field of rule extraction from neural networks. We also present a taxonomy for rule extraction techniques and highlight areas which we feel could be targeted for future research.

The aim of this research is to develop a training program tailored to individual abilities in the field of e-sports. E-sports require a diverse range of skills, and it is difficult to achieve significant results without training that aligns with an individual's capabilities. This study focuses on the FPS (First-Person Shooter) genre, evaluating the abilities and weaknesses of each player and developing a methodology to customize training programs based on this information. Specifically, the evaluation of abilities concentrates on crucial factors in FPS games such as reflexes and accuracy in hitting targets. The research aims to verify the effectiveness of the constructed training programs by implementing them based on the evaluation results. Moreover, data analysis and machine learning techniques are employed to collect performance data and training results from competitors. The goal is to discover effective training methods by leveraging these analytical approaches. Additionally, the research emphasizes the motivation and psychological aspects of competitors, aiming to develop supportive techniques that maximize the effectiveness of the training program.

The nature of the human mind is a long-standing question that has eluded definitive answers from varous scientists, but the truth has yet to be elucidated. One of the phenomena that has been studied as a sign of consciousness is the readiness potential (RP), which is observed prior to human voluntary actions. It has been found that RP occurs 500-800 ms a person intentionally acts, earlier than their conscious will for initiate the action. We conducted an experiment with different instructions for movement initiation and divided the time until movement onse into three phases: time of deciding (T1), time of planning movement (T2), and time of actual movement (T3). Participants performed tasks that required all of T1, T2, and T3, tasks that required only T2 and T3, and tasks that solely required T3. By comparing the onset times of each task, we found that T1 and T2 lasted approximately 25 ms, while T3 lasted approximately 360 ms. We observed a correlation between the duration of ”T1 and T2”, and a larger maximum amplitude of RP, indication that shorter T1 and a longer T2 were associated with larger RP amplitudes in movements similar to those in the present experiment. Moreover, we observed that task difficulty affected the maximum amplitude of RP, suggesting that harder tasks resulted in larger RP amplitudes and, therefore, shorter T1. This suggests the possibility that unconscious decision-making and preparatory processes related to task difficulty.

In recent years, mindfulness meditation has gained significant attention, particularly the use of Counting Breaths as an inducing method of meditation. This study aimed to investigate the efficacy of multi model stimuli in passive or active approach on inducing meditation and to identify the most suitable inducing methods of meditation for beginners. Evaluation indices of the meditative state include alpha and theta power of EEG in the frontal lobe, heart rate, and respiration rate during meditation. The quality of ”Concentration,” known to be improved by meditation,was assessed using the task performance and theta power of EEG in the frontal meditation.The results indicated that Counting Breaths (CB), Feeling Touches (FT), and Listening Sounds (LS) were effective in inducing both a meditative state and concentration. CB, FT, and LS showed a subjective increase in the feeling of ease for performing the task,and the physiological indices simultaneously indicated a meditative state. Passive sensory inputs (Watching Counting (WC), LS, Watching LED (WL), FT, Listening Counting Numbers (LCN)) or active attention (CB, Counting Watched Counting (CWC), Counting Listened Sounds (CLS), Counting Watched LED (CWL), Counting Felt Touches (CFT), Counting Listened Counting Numbers (CLCN)) had little impact on the magnitude of the meditation effect.

Otsuki had conducted categorical naming tasks on aphasia patients with categorical naming disorder, and reported a case in which the difference in performance was shown between "round objects" and "non-round objects", and showed that visual elements such as shape may have some influence on the recalling process [1]. Yamanoi et al. had presented images of a quadruped animal and measured EEGs from the subjects during recalling their names [4]. To these EEGs, the equivalent current dipole estimation was done. As a result, activities were estimated in areas related to visual cortex, memory, language cortex, etc., such as V1, hippocampus, and Broca's area. In images of typical animals with almost the same limb length, such as dogs and bears, brain activity was not estimated in the right angular gyrus, but brain activity was estimated in the right angular gyrus in atypical animals such as giraffes. In the study, the authors estimated the brain activity of the image recognition process of tetrapod image and compared with that of the recalling process. No activity of the right angular gyrus was observed in dogs in common with all subjects. For images of giraffe, brain activity was observed on the right angular gyrus in all subjects, and these tendencies were the same to those of the estimated brain activity in the recalling process. But activities were observed on the right angular gyrus for bear and lion in some subjects of the image recognition process.

In this paper, we propose a programming advising system based on organized knowledge related to programming problems found in textbooks. Specifically, when a user inputs an incomplete program, our proposed system identifies and predicts concepts that the beginner does not understand from their program code and provides advice based on these ungrasped concepts. We combine a large-scale language model with our organized frame knowledge to generate advice. Additionally, the proposed system enhances the learning experience by delivering advice in the form of textual explanations, coupled with an interface that visually displays error points.

To get qualified is an advantage when looking for work. In recent years, there are many tools that support studying for a qualifying examination. Reading aloud is effective for efficiently committing to memory, but there were no existing tools that use speech recognition technology to study for qualification exams. The purpose of this research is to develop learning support tools using speech recognition technology. The learning support tool to be developed uses voice for questioning and answering questions. We confirmed the effectiveness of the proposed tool by comparing the scores of subjects who studied the qualification test with and without the tool.

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 conventional system (editing image by changing each feature manually).

In Japan, we are suffering from increasing labor shortages. Tertiary sector such as restaurant tends to rely on part time employees. Currently, training in food preparation is often conducted through heavily OJT(On the Job Training), in which an instructor accompanies the trainee and provides guidance through actual cooking. However, the small number of full-time employees in the field makes OJTs training very difficult. To effectively instruct trainees, the trainer needs to grasp their mastery level of tasks, but it is very difficult to manage the mastery level of all trainees. In this research, we propose a support system for OJT in the form of simulated OJT using video images, aiming to provide appropriate training tasks according to the trainee’s situation and their level. This paper describes the design of the proposed system and our findings on the prototype.

We explored the influence of robot’s eye gaze on human behavior in interactions when passing each other. Nonverbal communication is important in human interactions, and eye contact in particular affects the way people keep their distance from each other. Furthermore, we devise a design of the robot look at people when passing each other. In addition, we devise to compare with the robot, a faceless robot with a rotating neck and a robot with a face without a rotating neck. In this study, experiments are conducted to the distance between the subject and the robot, and the subject’s eye gaze data of the subject. We observe that the robot gaze at the subject when the subject invades the robot’s perimeter (3.5 meter and 5.5 meter). The robot that gazed at the 5.5-meter perimeter was gazed at more continuously by the subjects than the other robots. 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 this paper, we examined the optimal class interval width for each user in order to auto- matically estimate the class interval width that determines the probability used in the DPTM continuous authentication algorithm. In the experiment, we performed the DPTM-based continuous authentication using multiple class interval widths, and examined whether the class interval width with the highest continuous authentication evaluation is different for each user. As a result, we found that the optimal class interval width differs for each user, indicating the importance of automatic estimation of the optimal class interval width.

In 2022, the number of fatalities and serious injuries in bicycle accidents with pedestrians will be 312. In order to prevent such accidents, we propose a bicycle-based collision warning system using the iPhone's LiDAR and object detection.

Perceptual system is one of the most important capabilities for an autonomous mobile robot in order to operate a task in unknown environment adaptively since the autonomous robot needs to detect the target object and estimate the pose of the target object for performing given tasks efficiently. In this paper, we introduce our design concept of Growing Neural Gas (GNG) based perceptual system for autonomous robots.

In recent years, Digitalization has become a hot topic, and in particular, the introduction of autonomous mobile robots is required in the service industry for safety and practicality. Therefore, we have been studying the traversability for autonomous robots and the degree of graspability of unknown objects using Growing Neural Gas (GNG) which is a one of unsupervised learning. However, previous methods extract pixels evenly from the input data for learning, which makes it difficult to collect information on the object of interest. In this paper, we propose a method for finding salient points in an environment using saliency which is a human gazing mechanism. Concretely, we have succeeded in obtaining the salient points of an unknown environment by dividing the features into color, intensity and orientation, and then integrating the salient points for each using GNG. Finally, we discuss the effectiveness of our proposed method using 2D images, 3D images, and benchmarks.

Recent research has shown that continuous learning models, which emulate the learning mechanisms of the human brain, can effectively learn new data over time. However, many of these models require significant amounts of task-specific data to extract visual and temporal features using convolutional neural networks (CNNs) and recurrent neural networks (RNNs). Additionally, pre-training these data feature extractors is often computationally expensive and time-consuming. Furthermore, traditional continuous learning models require data labeling before learning new data, which is difficult to achieve in real-world applications such as human action recognition. To address these limitations, we present a novel Random Weight Convolutional-Growing Memory Network (RWC-GMN) model that enables real-time continuous learning of new human actions in the real world. Our model uses a fixed random weight 3DCNN as a feature extractor, thereby eliminating the need for pre-training and enabling our model to start learning new actions immediately. The random weight features will be fed into the Growing Memory Network (GMN) for learning. GMN is a self-organizing incremental network that emulates human episodic memory. The network size can grow and shrink to adapt to input data continuously. Moreover, our model does not require pre-labeling of new human actions, allowing for label-free learning in real-time. Finally, our model can learn new actions continuously while retaining previously acquired knowledge, avoiding the problem of catastrophic forgetting.

AHP (Analytic Hierarchy Process) has been widely used in decision making and there are lot of extensions. However, using some AHPs, the results often lose reliability because the comparison matrix does not always have sufficient consistency. In this paper, we consider some fuzzy weights representation and aggregation for some extended AHP.

A human’s logical information processing seems simulated on the neural network in the brain. We trained 2-valued logical operations "and," "or" and "exclusive or" with a standard multi-layered neural network with the sigmoid function as activate function. We had similar results to those of fuzzy logical operations. In this paper we learn the "not" operation and show its results. The results of leaning depend on the activate functions. We propose a new activate function based on the S-function for defining fuzzy sets and show the results.

This study aims to accelerate iterative fuzzy modeling by proposing a method that computes only the errors in the region of interest. In this paper, we provide an explanation of the fuzzy modeling approach under consideration, introduce the improvement method, and demonstrate its rationale and effectiveness through experiments.

This paper discusses a method for identifying states in a multistage Decision Making Problem in which an Indifferent Event is either predetermined or can be automatically derived after the fact. First, when they are pre-set, the amount of possible information about Indifferent Event tends to be large. Therefore, since the decision is risk tolerant, the Max-Product method of Tanaka et al. is used to calculate the expected utility possibility. Next, in the case of automatic derivation after the fact, the amount of information on the possibility of Indifferent Event is relatively small, so the expected utility possibility is derived using Zadeh's Fuzzy Event Possibility Measure. Here, it is assumed that the setting of the utility function is independent of the information on the occurrence of the Indifferent Event and is identified by the decision maker by lot drawing using the certainty equivalence method. As a concrete example, we focus on the pass/fail decision of a recommendation test, which is a two choice question in the No-Data Problem, and illustrate the multistage state identification method.

Diabetes mellitus, one of the leading national diseases, requires constant control of blood glucose levels. Blood glucose levels must be kept within a standard range, and hypoglycemia, or low blood glucose, can be especially dangerous since it can lead to loss of consciousness and even death. In daily life, it is necessary to pay attention to the sugar content of food, inject insulin before meals, and take small sugar supplements such as chocolate or candy to adjust the blood glucose level. Moderate exercise is also important. The problem is that change in blood glucose levels and effect of insulin injection can be different for each person. Diabetes management, including blood glucose control, diet control, and exercise planning, must be tailored for each individual. To solve this problem, a personalized fuzzy decision-making system for diabetes management is proposed.