Observed color in images is easily affected by lighting conditions such as the sunlight location, weather, or time of day. Especially for image recognition processing on an outdoor navigated autonomous mobile robot, we should consider the effect of lighting conditions to achieve consistent robust person detection regardless of lighting condition change. The purpose of this study is to develop a system for detecting target persons by applying Convolutional Neural Networks. As a demonstration of person detecting capability for the proposed approach, we apply specified human finding tasks defined by Tsukuba Challenge 2021 rules. Using the proposed method, we were able to demonstrate that person can be stably detected and designated people can be judged regardless of surrounding light change.

Small ground vehicles, such as mobile robots, widely use a two-wheel steering mechanism that allows them to turn on the spot and move accurately and quickly. Without indicators to show how they move, such ground vehicles could collide with surrounding humans. Currently, however, it is difficult for mobile robots to communicate their behavioral intentions through nonverbal communication like humans do. Therefore, this paper examines the development of a new behavioral intention indicator in an environment where humans and robots pass each other in order to realize a society where pedestrians and robots coexist in urban areas in the future.

For autonomous mobile robots to travel safely, it is important for them to have a stable understanding of their surrounding environment. Most autonomous mobile robots use LiDAR (Light Detection and Ranging) to sense the surrounding environment. Under normal circumstances, this LiDAR can stably and accurately assess the surrounding environment. However, during rainy weather, it has a problem in that it recognizes raindrops as objects. When an autonomous mobile robot uses LiDAR to detect obstacles in rainy weather, the laser emitted from LiDAR may hit raindrops and misjudgment them. In this paper, we employ 3D-LiDAR and consider a method different from the conventionally employed clustering method to realize an autonomous mobile robot in a rainfall environment.

In recent years, educational support robots are paid attention to. The previous study reported the presuming perplexion method which presumes learner’s perplexion from the learner’s expression based on deep learning. However, the previous method presumes the perplexed state only from the learner ’s expression. Therefore, the method cannot presume when learner ’s expression cannot be recognized. If it is possible to construct a method which presumes learner ’s perplexion from body movement, the method will be utilized in modern society wearing a face mask is recommended. Thus, this study proposes the perplexion movement presumed method which presumes learner ’s perplexion from upper body based on deep learning. Moreover, we investigate the accuracy through simulation experiment.

In recent years, the STEM education has become active in public schools. Moreover, the programming education is becoming compulsory. Therefore, board games that learners can enjoy learning programming are being used in educational field. However, it is difficult for learners to gather partner who plays the board game at home. We think that the robots can replace the partner who plays board games together. If it can be realized, the possibilities of board games for programming education would expand. Thus, this paper investigate the impression and learning motivation that the robot gives to learners and examine the effectiveness of robots in board games for programming education.

Withdrawal among the elderly, which is one of the causes of the number of individuals requiring nursing care, has received much attention in recent years. A decline in self-efficacy is thought to be a cause. As a result, we believe it is critical to first assist them in understanding "what they can do," "what they cannot do," and "what they wish to be able to achieve." In this study, we built a measuring system that included three system components: a database management server, a sensor network system, and a robot system to track the movements of elderly people in a trailer-style living laboratory.

According to Brown et al. there are two basic human needs: the need for self-determination and the need to evaluate others, and language strategies based on these needs. They call them "negative politeness strategy (NPS)" and "positive politeness strategy (PPS)," respectively. The degree of face violation by a certain speech act is determined by "Distance", "Power" and "Rank of impositions". In other words, it is possible that a linguistic strategy appropriately selected in one culture may result in speech with low affinity in another culture, including differences in attitudes toward AI and anthropomorphic systems. In this study, we report the results of a survey conducted in six countries (Japan, the USA, China, France, the UK, and Australia) on the effects of linguistic considerations.

In general, fuzzy classifiers have high interpretability because fuzzy classifiers can linguistically explain the classification reason by fuzzy sets used in the antecedent conditions of rules. A reject option that rejects patterns near the boundaries between different classes is an approach to increase the reliability of fuzzy classifiers. However, the conventional threshold-based reject option may reject more patterns than necessary to achieve high reliability. In this paper, we propose a two-stage reject option where after the threshold-based decision, the k-nearest neighbor is used for patterns with low confidence value than the threshold. If the class labels predicted by the k-nearest neighbor and the fuzzy classifier are the same, the fuzzy classifier outputs the predicted class label without rejection. Through computational experiments, we discuss the relationship between accuracy and the rejection rate.

In recent years, pattern classification has been used for various real-world problems. However, there may be a bias toward certain attributes in data collection. This may result in inappropriate classification biased toward particular social groups. For example, when designing a classifier that recommends whether an applicant should be hired or not for recruitment problems, there is a possibility that attributes such as race and gender affect hiring outcomes. So far, we have developed multiobjective fuzzy genetics-based machine learning (MoFGBML) considering classification performance and interpretability. This paper incorporates two fairness measures into MoFGBML to design fuzzy classifiers considering classification performance, interpretability, and fairness.

The least squares problem (LSP) including general simultaneous linear equations is the most important model such as a linear regression and multivariate linear regression models in machine learning, as well as approximation problems in the material science, estimation of transfer functions in the control system, traffic flow prediction problem and various engineering optimization problems. It is attracting attention that the solution of this LSP model is obtained by iterative calculation and without using any inverse matrix. If the order of the LSP model is m, the conventional method by Yahagi's method required a 2m x 2 m matrix operation, whereas the modified Cholesky decomposition method proposed by Gen, et al has the advantage of the m x m matrix operation is sufficient with the small number of division operations. In this presentation, we propose a hybrid evolutionary algorithm (genetic algorithm and enhanced Jaya algorithm: HGA+EJA) for effectively solving LSP models including nonlinear LSP models including general simultaneous linear equations. In the numerical experiments, several numerical examples of general simultaneous linear equations, the ill-structured simultaneous linear equations, and linear/nonlinear LSP models are demonstrated, and the solution precision and computational time by the conventional method and the proposed method: HGA+EJA are quantitatively compared with each other. Finally, we clarify the effectiveness of the proposed method.

Reinforcement learning is often used to solve AGV problems, such as the package transport problem. Roulette selection and epsilon-greedy are typical strategies for action selection in reinforcement learning. In this study, we examine the transport efficiency of AGV for the two types of action selection and verify a method that combines the two types of action selection.

In recent years, a decrease in the number of farmers and the aging of the farming population in Japan have led to a labor shortage and a decline in production efficiency. One solution is the use of artificial intelligence technology to improve work efficiency, but this requires a large amount of training data to be developed. In this study, the goal is to develop a clustering instrument to select scratches and corrosion of crops, but it is difficult to obtain a large amount of data on scratches and corrosion. In this paper, we report on our efforts to create a variety of data using Cycle GAN by interpolation techniques, while maintaining the required number of data by using the already obtained data.

JADE is an optimization algorithm that uses probability distributions to adaptively select parameters. However, it does not take into account the search for regions outside the solution population, so it can be improved by adding an efficient out-of-population search such as the Nelder-Mead method. In this study, a simple method with a small number of parameters to add out-of-population search was considered while keeping the search speed as high as possible, and its effectiveness was confirmed through numerical experiments.

Many studies for privacy preserving data mining have been done. Secure Multiparty Computation (SMC) has been introduced for privacy preserving data. SMC was applied to some statistical computation methods. In previous papers, we proposed synchronous Back Propagation (BP) learning for SMC. In this paper, we propose the improved BP learning method with small number of parameters, even if the number of data and weight divisions is large.

We observe the brain activity of human subjects when they are exposed to image stimuli using fMRI. We extract feature maps of the same images using Attention Branch Network (ABN) and construct a regression model that captures the correspondence between the enhanced image features and the brain activity data. This regression model will be used to examine how weighting features by ABN rather than image features extracted by CNN works in estimating the brain activity state.

By calculating time-series correlations of EEG data between brain regions, we can obtain the weighted graph containing the nodes of each region, which allows us to express the functional connectivity (FC) of the brain. Recently, methods for analyzing brain states in terms of such brain networks have attracted increasing attention. Studies have often focused on a single network constructed from data over a period of time, however the FC actually changes over time. This study proposes new time-series features obtained from the moving correlations of EEG. We compared the classification accuracy of the task estimation between normal and moving correlations.

False memory is a phenomenon of ”remembering what has not happened” or ”remembering what quite differently from the way they happened ”. In many studies with DRM paradigm which easily generates this phenomenon, the experimental participants perform the cognitive tasks alone. In this study, we analyzed an influence of relationships with others in conversation on the generation of false memories. Participants in this experiment memorized words selected based on the DRM paradigm and shared the information about the words with the agent with text-to-speech software through an online conferencing system. We analyzed the influence of explicit information on the reliability of the agent by indicating the reliability in % of the agent, the influence of the impression factor of others by changing the gender of the text-to-speech software, and the influence of confirming the relationship through courtesy by instructing the participants to bow to the agent. As a result of the experiment, the rate of false memory increased when the explicit reliability level matched the perceived reliability level of the participants, and the rate of false memory generation for male agents was higher than for female one. The results also indicated that the rate of false memory generation increased when the participants performed polite behavior. Less activation of working memory activity was observed, suggesting that the interaction with an agent has inhibitory effects on the generation of false memories.

The right angular gyrus has been associated with visual attention toward distinctive features. One of the authors, Mika Otsuki, reported an example of a naming task using line drawing of fruit or animal for aphasia patients, and showed a difference in percentage of correct answer between "round things" and "not round things". In a continued study, some of the present authors had analyzed spatiotemporal human brain activities during the tetrapodal image recalling process by use of the equivalent current dipole source localization (ECDL) method, and found that activation of the right angular gyrus played an important role. Same as the previous study, the present authors had analyzed brain activities in image evoked recognition process of four animals, and compared the results between two processes, and found also the right angular gyrus played a role on shape discrimination between abnormal and normal (round), but the results are slightly different in subjects and images. And the authors also analyzed the single-trial electroencephalograms on the first image presentation of each image in order to verify whether comparison process with the memory involved.

Analytic hierarchy process (AHP) models mainly have been used for calculating the global evaluation values and the aids of decision making. Using the Choquet integral calculation process,evaluation values of the alternatives are assigned as a set-function representation. The set-functions of are aggregated to the mean set-function. As the mean set-function representations of examinees hold the rank information of each examine evaluation values, the mean set functions are used for the analysis of alternatives. AHP set-functions of examinees can analyze various situations such as consensus, indifference, multi-polarization, passable alternatives, error, and conflict.

In this paper, interval priority weight estimation methods from a pairwise comparison matrix are studied. This estimation problem has non-unique solutions except for special cases. The usefulness of the interval priority estimation has been investigated by demonstrating the accuracy in ranking alternatives by the estimated interval priority weights whose sum of the centers is one. In this paper, we investigate the accuracy of the set of estimated priority intervals. As it is not very easy to evaluate the accuracy of the whole members of the solution set, the accuracies of both endpoints of the bounded interval composed of whole solutions in the vector space are evaluated together with the accuracy of the estimated interval priority weights whose sum of the centers is one in this paper. The results of numerical experiments are reported.

Fuzzy classifiers can linguistically explain its classification mechanism while achieving high classification accuracy. In this paper, we aim to explain the classification mechanism in dynamic environments where the classification boundary changes over time. For this purpose, a fuzzy classifier is constructed in an online manner by means of Confidence-Weighted learning. Online learning allows the classifier to be trained from a small number of training patterns. We have confirmed that the learning model can linguistically explain the classification mechanism by examining how the weights of the fuzzy if-then rules in the fuzzy classifier so that the fuzzy classifier dynamically follows the changes in the classification boundary.

This paper proposes an adaptive Monte Carlo method based on Exchange Monte Carlo and gives a discussion with respect to it through experiments. Exchange Monte Carlo is used widely for both purposes of optimization and estimation of a probability distribution. There have been studies to propose an improvement on the method, however, adaptive method has not been studied well. The paper proposes generalized joint probability of parameters’ distribution, and target distribution that is desired to be estimated. Using Hamiltonian Dynamics in addition, the proposed method approximates the original distribution. Experiments show the adaptive functionality can be observed, and original distribution is approximated well.

In recent years, severe economic damages have been caused by the spread of the Covid-19 infection. The design of economic support policies using social simulations has been paid great attention. In the Evolutionary Computation Competition 2021, a multiobjective optimization problem was posed to design economic support policies that simultaneously optimize the elimination of impoverished conditions and an appropriate level of payments. Economic support policies are evaluated using a social simulation with synthetic population data and multiple economic shock scenarios. In this paper, we apply an evolutionary multiobjective optimization algorithm to the design of promising economic support policies. We analyze the characteristics of Pareto optimal economic support policies and the structure of the multiobjective optimization problem through computational experiments.

The average life expectancy of human beings is increasing year by year due to advances in science and technology, and while it is said that the era of 120-year life expectancy will eventually arrive, the decline in cognitive and motor skills and the shortage of human resources for caregiving have become problems. On the other hand, in the ultra-smart society of Society 5.0 proposed by the Cabinet Office, it is expected that "coexistence between humans and robots". In this way, in the era of 120 years of life, human life support by robots is important. Based on this background, we have constructed a trailer-type living laboratory and are conducting research and development of a life support system that uses stationary sensors and mobile robots to measure human daily life activities, estimate intentions, and support, maintain, and improve behavior. However, this doesn't have system that can collect, accumulate, and analyze from the history of robot and human actions. In this paper, we discuss the effectiveness of this system in terms of real-time performance by predicting movement based on the phase structure of movement history as an example of analysis.

Interior image recommendations have been paid attention to with the development of image retrieval technology. Although a classification model is often used to model user-preferred interiors, it is not suitable for explaining why users prefer them. On the other hand, LDA, which analyzes sentence features based on latent features, can provide a better understanding of sentences by estimating the latent features in the sentences. Therefore, in order to understand the structure of preferred interior images by users, we propose a method to estimate latent features in interior images by incorporating spatial features of images and LDA. Specifically, instead of sentences, we use interior images to extract spatial features using BoVW based on SURF features and their histograms, and estimate topics based on the histograms by LDA. We compared the proposed method with subject experiments in interior images with several styles and verified that the proposed method can estimate similar interior topics compared to the subject ’ s subjective view.

In this study, we performed the robot lectures in which a robot as a teacher interacts with participants. There are many participants who answer the robot questions, but there are few participants who ask questions. The reason can be found in what participants are daunted by being worried about the opinions of others and hesitate to ask questions. For sharing the comments or questions, we propose that all participants can see the questions of others and the robot reads them out. To relieve hesitation, we propose that the robot posts shill questions to look like some participants ask questions. Furthermore, we develop a system that allows participants to easily send their opinions by pushing buttons down to agree with comments of others. As an experiment, we perform two demonstrations of robot lectures. We discuss the effectiveness of the proposed robot system.

In this study, we investigate the impression that the robot’s refraining in the virtual space gives to the human operators of the agents in the virtual space. Research on social communication robots is underway toward the realization of a human–robot symbiosis society. In such a society, robots should behave cooperatively when interacting with surrounding humans. In order to realize human–robot cooperation, we have been developing robots that can adapt to implicitly formed group norms even without direct interactions. In this study, we focus on a new aspect of implicit cooperation between humans and robots: “refraining,” which means withholding action in relation to others. We investigate the impression of a robot that expresses refraining behavior according to the behavior of humans who are engaged in a cooperative task with the robot in a virtual space. The results suggest that the robot’s refraining behavior promotes the human’s active behavior and suggests the construction of a relationship in which the robot and the human support each other.

We propose an assist control for a motor-driven powered suit for lower limb assistance based on head motion detection. The assist control system detects the head motion of the wearer in lifting motion using a 9-axis inertial sensor and controls the powered suite based on the motion detection. The effectiveness of the proposed method was verified by electromyography. As a result, the EMG of the vastus lateralis muscle decreased, confirming the assist effect of the exoskeleton.

The mini 4wd is commercially available as a children’s toy. There are many toys that adults have modified. This paper captures the mechanical aspects and how to actually improve it.

In this paper, we describe a method for automatically controlling the speed of a mini 4WD car through the localization using image processing. A camera captures images of a mini 4WD car in motion, and its position is estimated from the difference between the background and the captured images. The mini 4WD car is then controlled based on the position of the mini 4WD car and the control rules that are specified in advance. There is a communication delay during the control of the mini 4WD car. This problem is worked around by setting the control rules so that the communication delay is taken into account when the control rules are specified.

In team sports with a high degree of freedom of movement, players and referees on the field need to make immediate decisions on movement actions that maximize team and individual gains. To do this, players and referees are constantly performing the task of quickly recognizing the game situation. Because the human field of vision is limited, they must select and focus their attention on the information they need to pay attention to. This kind of behavior to actively gather information about the surroundings is called visual search behavior. However, it is difficult to specify what information to pay attention to, and it is not easy to measure and collect visual search behavior in actual games. In this research, we developed a system to measure visual search and movement behaviors in a virtual space for efficient data collection and analysis. Using a head-mounted display (HMD) with a built-in eye tracking device, the system enables measurement in a more realistic environment by immersing the user in a match reproduced in a virtual space by a soccer simulator. We conducted an experiment in which testees were immersed in a match in a virtual space from the viewpoint of a referee, and collected behavioral data as a referee. As a result, we confirmed that there were significant differences in behavior depending on the presence or absence of knowledge of soccer and experience as a referee.

In recent years, online classes have rapidly increased due to social conditions such as COVID-19. This has led to the accumulation of a large amount of evidence on learning in learning management systems (LMS), and online classes have become widespread. However, there are some disadvantages, and students are sometimes left behind. In this study, we analyzed and compared educational data using two machine learning models to examine the differences in contributing features.

Teachers make student groups based on their profiles for effective collaborative learning in the classroom or online. We represent a student profile with their inner evaluations of both knowledge level and social interaction criteria. Each student evaluates herself on how much confidence s/he has in each criterion from the optimistic viewpoint. At the same time, from the pessimistic viewpoint, s/he gives self-evaluations on how much worried about the criteria. The peer evaluations from the other students replace the self-evaluations on confidence and worry. The judgments from several viewpoints increase reliability, though the multiple perspectives are not always consistent. The inconsistency stems from the uncertainty of the actual evaluations. In this study, we denote the student profile as a normalized interval vector of the criteria evaluation to reflect the uncertainty. We propose the model to obtain the interval evaluations of criteria from those of multiple crisp evaluations considering relative relations. The idea is similar to Interval AHP, which derives the interval normalized vector from a crisp pairwise comparison matrix. Then, we form the groups so that their group performances are similar.

A large number of high-strength bolts are used to connect multiple steel members when erecting steel bridges. All high-strength bolts are visually checked one by one to confirm whether they are tightened correctly or not, which requires large effort. This paper proposes an automatic bolt detection method to automatically determine whether the high-strength bolt is tightened correctly or not. In the proposed method, images of multiple high-strength bolts are raster-scanned, and a presence probability map is created from the HOG features of each small region and the high-strength bolt detection frame obtained from SVM. Since the bolt size is not constant in the image, the exact bolt bounding box is estimated by fuzzy inference using the obtained detection frames and existence probability values. The accuracy of the proposed method was evaluated using 42 images taken at a steel bridge erection site. Experiments results showed that the proposed method can detect high-strength bolts in 0.753 precision, 0.614 recall, and 0.676 F-measure with fuzzy inference.

A human’s logical information processing seems simulated on the neural network of brain. We have trained a neural network with 2-valued logical operations "and" and "or" with a multi-layered neural network and we have learned fuzzy operations like the bounded-product and bounded-sum, where the neurons have the sigmoid function as activate function. In this paper we learn the "exclusive or" operation. We have two patterns of results and discuss the result.

It is known that human visual system has the mechanism of figure-ground separation distinguishing foreground object and background regions. The border-ownership coding, represented by neurons having the selectivity for figural side against contour, in addition to orientation selectivity, is thought as the internal representation of the figure-ground relation in the brain. However, we usually look 3D environment binocularly, and the boundary of objects is 3D surface, rather than 2D contour. It is necessary to extend the nature of the border-ownership from 2D to 3D, where each local surface patch belongs to one side between two sides of adjacent subspaces. How such information processing is realized comes to a problem. This study presents a neural network model assigning the ownership of 3D surface. We focus on the nature that alike 2D object’s boundaries as contours, 3D object’s boundaries as surfaces have globally convex shapes, though objects have generally both convex and concave parts. We extended 2D figure-ground separation model proposed by Kikuchi and Akashi (2001) into 3D model with similar principle. We supposed the signal propagation by local averaging equipped in each two sides against the surface, and mutual inhibition between the two sides. Initial values in proportion to the magnitude of curve is given to the inner side of the curve. We confirmed that through the iterative process signals remains in only one side corresponding to the inner region of object.

Synaptic plasticity in neuronal networks has been shown to be important for motor learning. In this study, we analyzed the neuronal network that interacts with the outside world through the robot body to induce collision avoidance behavior of this neuro-robot, by reducing the number of spikes in spontaneous electrical activity due to the short-term depression induced by repeated electrical inputs to the cultured neuronal network. When the robot detects an obstacle, it is designed to avoid collision with the obstacle by slowing down its motor speed in response to the short-term depression (STD) expressed in a subset region of the network, induced by continuous inputs to a certain stimulatiuon electrode corresponding to the position of the obstacle. We analyzed the STD effects by the stimulus patterns and performed a running experiment of the neurorobot with the optimal stimulus condition (10 Hz). As a result, the robot succeeded in expressing the collision avoidance behavior. It was observed that the turning behavior increased. The spike frequency of spontaneous electrical activity before and after the running experiment tended to decrease, and the periodic intervals between spike firings tended to be larger than those before the experiment. These results suggest that the running of the neurorobot with avoiding obstacles induces long-term depression (LTD) in the cultured neural network and increases the interval width of the firing cycle of spontaneous neuroelectric activity.

In recent years, attempts have been made to elucidate human consciousness, and implementation of consciousness on machines have not yet been realized. In this study, we attempted to develop a system that uses inputs from the outside world as information to generate a primitive consciousness of the living neural network. The in silico system, a preliminary step using a living neuronal network, mimics the process of language acquisition process in infant period. We think that the simple process of segmenting words can be modeled by the mutual-segmentation-hypothesis, the system forms a semantic network by associating the segmented words with the current internal state considering the history of input stimuli, based on the mutual-segmentation-hypothesis. We analyzed the learning results in the case of sentences consisting of the simple words as inputs, and found that some of the words in the semantic network were correctly learned, but some of the words in the output were not natural, because these words were output repeatedly. Furthermore, we analyzed the characteristics of I/O of the living neuronal network, in order to utilize a living neuronal network as the process of changing the internal states by input stimulation.

Recently, we have been increasing interest in ensemble learning. In particular, Boosting is a useful learning method in which multiple classifiers construct multiple layers with maintaining mutual dependencies, and its expressive ability is high. We have proposed pdi-Boosting, which virtually generates interpolated data and adds them to the training data to improve the accuracy of classifiers. However, in this method, the area for generating the additional fuzzy rule is not directly related to the misidentification data and virtual data. Therefore, the discrimination rate is not high. In this paper, we propose a new \pdi-BoostingG" that also regularizes the additional region G of fuzzy rules. In pdi-BoostingG, fuzzy rules are additionally generated around the misidentification data, and virtual data is generated in this area G. In addition, the individual learning of the additional fuzzy rule and the whole learning of the whole space are alternately learned to form the multi-layered structure. As a result, pdi-BoostingG improves the discrimination rate and robustness. In addition, since virtual data and fuzzy rules are inherited between multiple layers, deep inference of fuzzy rules is realized. We formulate here how to generate additional fuzzy rules and virtual data, and how to inherit whole fuzzy rules and virtual data between layers. In addition, we compare the discrimination rate of pdi-BoostingG with other methods in order to discuss its usefulness.

In recent years, the shortage of childcare workers and the increase in the number of children waiting for admission to preschool due to an increase in the number of children who wish to enter preschool as a result of dual employment have become social problems. One of the reasons for the shortage of childcare workers is the sheer volume of work. If we could develop a methodology to estimate the characteristics of each child, we could reduce the workload of childcare workers and adopt a more efficient training policy. In this study, we modeled a series of behaviors of infants in group conversations by reinforcement learning. We also discussed the characteristics of each child by visualizing the estimated values of parameters and the learning process obtained from the behavioral data of each child

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 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. Perform text mining analysis on the collected data. The purpose is to extract useful knowledge and knowledge based on the results. We focused on Twitter, which has high breaking news and diffusivity among SNS. Twitter can easily send information and is considered to be suitable for obtaining various knowledge. The purpose of this study is to obtain knowledge about tourism by collecting data posted on Twitter and extracting data related to tourism from it.

We conducted a questionnaire survey (written in Kansai Dialect) on the self-evaluation of the acquisition of interpersonal abilities other than Essential Competencies for the 100-year life, which the Japanese METI defined, by a business major group of current students and graduates of Kindai University through PBL activities under the same seminar. From the survey results, we will consider the clues that will help us infer what skills are necessary to survive in the unpredictable era and examine how to prepare new PBL activities in the post-COVID-19 area.

In this paper, as an analysis of financial indexes we describe consideration on the classification with financial indicators based on the correlation.

With the development of economy and society, the utilization of the Internet in all walks of life is booming, and corporate social responsibility (CSR) has also evolved from a traditional offline model to a combination of online and offline models. The use of social media for value co-creation is called virtual value co-creation. China’s Alipay platform launched a public welfare activity Ant Forest, which is an example of virtual value co-creation. Behind this seemingly simple activity is a complex value co-creation behavior. This article considers Ant Forest as an example, guided by the theory of value co-creation, and selects the post-90s as the research object to explore the motivation of users to participate in virtual value co-creation.

Even-sized clustering based on optimization (ECBO) is a clustering algorithm that classifies the cluster size to be the same. It has been applied to problems such as K-anonymization in information security and transportation area segmentation in package delivery planning. However, ECBO may not provide the best classification when the cluster size does not need to be exactly equal and when some margin is allowed. Fuzzified Even-Sized Clustering Based on Optimization (FECBO) is a clustering algorithm that considered to be one of the solutions to this problem. By the way, data in the real world includes uncertainties such as having errors, widths, missing some of the attributes, etc., so we should think in terms of ranges, not points. However, FECBO is not able to handle data with tolerance. In this paper, we propose a new clustering method that extends the objective function in FECBO to the case where the objective function is a more general nonlinear function, such as entropy regularized one, and introduce the concept of tolerance to FECBO to handle those.

There are some variants of fuzzification technique in Fuzzy c-means, such as Bezdek-type, KL-divergence-regularized type, q-divergence-based type, and some fuzzy clustering algorithms with di- mensionality reduction methods have been proposed. However, not all these combinations have been proposed. There is a potential to increase clustering accuracy by variously combinig a fuzzification tech- nique and a dimensionality reduction method. In this report, eleven fuzzy clustering algorithms are pro- posed based on some dimensionality reduction methods, and the proposed methods are compared in terms of accuracy.

By using the clustering results of the Fuzzy c-means (FCM) method for cancer gene data, the discrimination results obtained by applying Support-Vector Machine (SVM) to each cluster, and the discrimination results based on the clustering results of the K-means method are compared. As a result, it was suggested that the results obtained by the FCM tended to have less variation in discrimination performance among clusters than the results following the K-means method. Originally, it was said that the types of cancer, based on gene expression data, can reflect more actual data if treated as a complex group with various labels of the types of cancer. The performance evaluation of this study shows that the extraction of a fuzzy cluster with multiple properties applies to identifying the types of cancer from this type of data.

While the usefulness of AI goes without saying, the development of reliable AI construction methods is essential to further accelerate the implementation of AI in society. For example, AI in the medical field is often used in primary screening to reliably exclude the majority of normal samples, and a high degree of trade-off is expected in making more error-free judgments with more samples. In this report, we introduce a new parameter to the loss function of the Siamese network of the deep distance learning model to enable mapping of hard-to-judgment samples to locations that reflect ambiguity among classes for the image classification problem. We also show that the mapping by introducing the new parameter provides a range in which samples that can be classified reliably can be extracted.

In clustering, data are classified based on similarity which is a measure of closeness between a part of data. Similarity is often expressed as a value in the interval [0, 1] rather than as a discrete value represented by {0, 1}, because it is expected to provide fine-grained classification, high versatility, technological development, and theoretical depth. In particular, probabilistic similarity is theoretically interesting because it is a measure that evaluates similarity not by its own value but by the probability that the similarity takes a predefined value. However, clustering and pattern classification methods based on probabilistic similarity have not been discussed much. In this paper, we propose a new hierarchical clustering algorithm based on probabilistic similarity and verify its effectiveness.

Space debris is artifacts that are dumped into orbit at the end of satellite operations or at the time of rocket launch, and is increasing in number as space exploration becomes more active. Since space debris poses a risk of collision in orbit during the launch of new spacecraft, studies on debris disposal have been active. In addition, orbit design is required so that debris can be disposed of at sea, avoiding land, where unburned debris can cause human and material damage due to the possibility of falling debris. In particular, since space debris has lost its own control capabilities, it is important to optimize the transition of the orbit up to the time of re-entry. Since the trajectory design problem is highly nonlinear, the application of metaheuristics is promising. In this study, we present the results of an attempt to minimize the hazard by using evolutionary computation to evaluate the ground hazard (distance to coastlines) from the orbit transition to the arrival at the earth’s surface through simulation.