Recently, automated negotiation has been attracting attention in multi-agent systems to resolve conflicts and reach an agreement among agents. In automated negotiation, two main types of strategies are incorporated in each agent: a bidding strategy that considers what kind of bid to send to an opponent, and an acceptance strategy that considers whether to accept the opponent's offer. In most bilateral multi-issue negotiation, agents take turns sending bids to each other and the negotiation ends when an agent accepts an opponent's offer. Therefore, the acceptance strategy is important in terms of increasing the utility of an agent. However, most studies of automated negotiation using reinforcement learning focus only on the bidding strategy of the agent, so there are not many studies that investigate acceptance strategies using reinforcement learning. In this paper, we propose a new configuration of a deep reinforcement learning framework for the acceptance strategy in automated negotiations using Deep Q-Network. The training phase is performed multiple times with various reward functions, and the reward capable of a higher utility value is investigated. Simulation experiments with other negotiating agents showed that the proposed method obtained significantly higher utility values than existing methods.

Agent-based computational economics has been attracting attention as a method in recent years; however, its model structures tend to become complex due to the coexistence of systems for decision-making and those involved in financial transactions. This means that a large amount of effort is required at each stage of the research. This study uses accounting-related concepts to develop a framework that extends bookkeeping notation and functionality to express the relationships between monetary entities in the model, which improves the working efficiency of Agent-based computational economics research. A case study of actual model construction is presented. The introduction of accounting-related systems can enable the separation of decision-making interaction systems from those involved in financial transactions to clearly express without misinterpretation, and “who performs what actions to whom and how such actions affect the state variables of each entity.” This improves the effectiveness of the model design, programming, and simulation phases and enables descriptions that are less likely to be misinterpreted by the reader when the results are published.

Ransomware attacks targeting Network Attached Storage (NAS) devices have occurred steadily in the threat landscape since 2019. Early research has analyzed the functionality of IoT ransomware binaries but failed to reveal its operation and attack infrastructure. In this paper, we propose an attack observation system named SPOT, which uses popular bare metal NAS devices, QNAP, as the honeypot and the malware sandbox to conduct an in-depth analysis of IoT ransomware attacks. During the six-month observation from September 2021 to March 2022, we observed on average, 130 hosts per day accessing from the Internet to compromise the NAS devices. Moreover, we executed 48 ransomware samples downloaded from VirusTotal in the SPOT sandbox. We identified seven remote Onion proxy servers used for C&C connection and successfully observed three samples infecting the NAS device to connect them to the C&C server behind the TOR network. The ransom notes gave two kinds of contact points; instruction web pages and email addresses. Though the email addresses were not reachable, we could access the instruction website. We kept monitoring the website and observed a “30% discount campaign” for ransom payments. We also interacted with the threat actor via online support chat on the website, but we were banned from the channel because we asked about their organization. We observe that the degree of automation in the attack operation is much higher compared to the carefully tailored and targeted ransomware attacks. While each case of successful ransom payment is limited to 0.03 BTC, the automated nature of the attacks would maximize the frequency of such successful cases.

Conversational agents are becoming popular in various daily situations. To make agents more user-friendly in the future, we are studying ways to make them more humorous. Our proposed interaction style for the agents is to perform boke (performing jokes) and tsukkomi (butting to jokes) based on Japanese manzai during communication with the user. To improve this style, we focus on the prosodic features of the agent's voice which affect the user's impression of a agent. The results of the experiment suggest that the user's sense of humor, friendliness, and motivation to continue the conversation may be improved by making the agent's voice speed normal and the pitch high. Also, it was suggested that the faster speed for the agent's voice may not be suitable for improving the user's sense of humor, familiarity, and motivation to continue the conversation.

Anomaly detection is the main topic in artificial intelligence and a crucial factor in productivity. The anomaly detection model based on generative models is a prime approach in this field, such as AnoGAN, CBiGAN. However, most of the current anomaly detection models with generative models are not accurate enough to reconstruct images. These cause differences between the detection image and reconstruction image even in normal regions of the detection image, which seriously affects the detection accuracy. To solve this problem, this paper proposes a new method for anomaly detection called ACGan. It uses CBiGAN as the generative model and adds an attention network based on the U-Net structure to find anomalous regions in images. The method can avoid errors caused by the lack of accuracy in reconstructing images by focusing the model's attention on anomalous regions. In this paper, three training methods, unsupervised learning, supervised learning and supervised learning with noise data, are designed. Experiments on the realistic dataset MVTec AD validated the effectiveness of the model. For unsupervised learning, the model has higher accuracy than CBiGAN on most product images. The models trained by supervised learning with noise data are highly robust. And the model has super high accuracy and is adequate for practical industrial needs if supervised learning is used.

Parents and guardians were supporting their children's learning prior to Covid-19, but evidence has indicated that many were not prepared for what happened and what was required of them during the Covid-19 period. Uses of digital technologies to ensure continuity of learning was frequently and rapidly used and developed, and pertinent literature often indicated the challenges that parents and guardians found in supporting their children through uses of these media. In moving from the previous practices to those needed in the Covid-19 pandemic time, this paper asks why parents and guardians were not prepared: what features were missing, and how might parents and guardians be better supported during periods of future crisis?

Science, technology, engineering and mathematics (STEM) is an emerging field of research within early childhood education and care across the world. It aligns with approaches to support early learning and skill development, given the recognition of the importance of the first five years of life. In this study we explored the framework documents in Australian early childhood education to understand more about STEM in early childhood education and care. Our study is interested in the frequency of STEM, as well as the meanings associated with STEM with regards to child learning. We engage with a content analysis, looking at latent as well as manifest coding. Key themes emerge associated with the absence of some STEM areas, compared to others, as well as the positioning of STEM learning areas. More specifically, digital technologies dominate the other three disciplines of STEM. It is anticipated that by providing policy understanding, conversations can commence to support the implementation of STEM in early childhood education and care.

Programming is a creative activity, but it can be difficult to learn due to constant updates, poorly maintained documentation, and unexpected errors. One reason for the difficulties is the shortage of programming teachers, which often leaves students unable to get help when they need it, even for simple questions. Many unanswered questions are a barrier to improving programming skills for creative purposes. The purpose of this paper is to address this issue by exploring whether an AI-based system can help reduce the difficulties faced by students. Recent advancements in deep learning technology have made it easier for teachers to train AI models that can learn from their own experiences in the classroom, including the types of questions, requests, and difficulties that students encounter. We have developed an AI model that can translate Python code from Japanese by using machine translation techniques and large language models. We have integrated this model into a learning assistant system that suggests code to students when they express their programming intentions. In this paper, we present our experiences in developing and deploying this AI-based assistant in the classroom, as well as the feedback we have received from students. By sharing our initial experiences, we aim to envision the potential of educational AI development for the future.

The situation after pandemic school closures has revealed drastic global inequalities and disparities in online education. With new efforts to improve the situation, the constraints and risks that are incurred in the digital transformation of education need attention. This paper explores and analyzes the risk areas dividing them into technological infrastructure, hardware systems, and software and platform concerns, and to wider social, ethical, political, epistemological and cultural issues. A systematic analysis will help in risk evaluation for better policy making and planning in the global south. The paper argues that technological transitions need support in the form of considerable expertise and involvement of the communities to ensure sustainable solutions in education. This paper also seeks to contribute to a broader discussion on the critical research agenda in development and technology studies in education.

Due to the influence of COVID-19, there is a growing need to implement online classes in junior high and high schools. In online classes, students generally use a mouse and a keyboard to interact with the PC, but there is a concern that using a mouse and a keyboard may cause students to lose concentration because they do not move their bodies. Therefore, in this study, we developed a large button corresponding to Zoom's reaction function. Because students need to move their bodies a lot to press this button, students will be able to maintain their concentration and enjoy the lesson. Using the developed button, we experimented with first-year junior high school students. We confirmed through questionnaires that using the button helps them to enjoy the class and concentrate more than usual.

In line with the global trend, programming education became compulsory in Japanese elementary schools in 2020. This study conducted an online questionnaire survey of elementary school teachers nationwide. A mixed methods research was attempted by combining quantitative (chi-square test) and qualitative (case-code matrix) methods. The responses collected from across the country were compared with a previous survey conducted using the same methods in 2017, which revealed the following: (1) ICT use is advancing, and informatization, or DX in education, is progressing in both schoolwork and teaching/learning slowly and steadily. Child-centered teaching and learning methods using ICT in classrooms are spreading. (2) Although the percentage of respondents who have implemented programming education has increased significantly, half of the respondents have never taught programming. While the diversification of programming education practices is evident, the author also found that programming education is highly dependent on external environments such as technology developers. (3) While confidence in teaching programming increased significantly, confidence in using ICT decreased significantly. The rapid transition to DX is considered to be increasing the workload of teachers. (4) The physical environment for programming education is considered to be improving; however, some teachers remain skeptical of programming education.

Visual programming languages are popular because they can be learned at an early age. Unlike text-based languages, they do not require linguistic knowledge. Visual programming languages are diverse and include block programming, flow programming, AR-based programming, and robot-based programming. Scratch and Viscuit are common visual programming tools. Because effective learning methods for these languages have yet to be clarified, we proposed a method and evaluated its learning effectiveness using Chuggington Programming, which is a visual programming game application. Although the questionnaire responses for five workshops at educational institutions were positive, quiz results for one- to two-hour workshops did not show significant differences based on Wilcoxon's signed rank test. Herein we focus on complexity in visual programming to assess the feasibility of a new evaluation method similar to a text-based evaluation method using turtle graphics. We compare traditional text-based evaluation methods such as Cyclomatic Complexity, Halstead Complexity, etc. with turtle graphics. Then we examine the relationship between data for student performance in the workshop and the complexity level. The solution time and complexity are positively correlated, while the percentage of correct responses and complexity are negatively correlated, validating the complexity has potential as an effective learning measure.

Since FY 2020, programming has been a mandatory subject in elementary schools in Japan. Each elementary school decides the grades and subjects for programming in each aspect of curriculum management. However, there have been few reports on school-wide programming education. In this paper, we report on school-wide programming education delivered at an elementary school since 2017. We also report the content of the instructional plans and lesson practices for each grade level developed based on five years of classroom practice. In addition to this, a comparison of the results of the responses to the 2018 and 2021 questionnaires completed by enrolled teachers is used to examine changes in teacher awareness due to school-wide programming education initiatives. Questionnaire surveys of the pupils and teachers suggested that the lesson practices had elicited the pupils' interest in programming and enhanced their subject knowledge. It was also suggested that the teachers' knowledge and awareness of programming education improved through continuous school-wide efforts.

We have analyzed what students think about e-learning and face-to-face classes during COVID-19 by using text-mining. In the spring semester of 2020 instructors and students experienced for the first time class tuition that was not face-to-face but e-learning. These students showed significantly different features compared to students in other semesters.

Computer Science Unplugged was originally designed to teach children in a face-to-face environment. However, the author has adopted it for teaching in an university class. While it was previously conducted in face-to-face manner, after COVID-19 began, the author modified the program to suit the online environment and used it in teaching online. The author discusses the rearranged Computer Science Unplugged activities for the online environment and presents the result of teaching them to undergraduate students.

We conducted programming classes in the second semester of 2021 at an engineering college. These classes targeted first-year students in online settings. A total of 137 students from the Faculty of Engineering participated. Classes were delivered in real-time and students had the option to watch the recorded videos and submit their assignments on demand. This setting initially presents two potential problems: 1) Teachers cannot ensure students participate on time; 2) Unmotivated students might opt for on-demand study and eventually neglect their studies. To address these anticipated disadvantages, we implemented a strategy where students had to write reaction papers and submit assignments after each session. Through this approach, we aimed to sustain their motivation. We conducted four courses. Although three of them had a high number of real-time attendees, the scores showed no significant variations. The questionnaire associated with the reaction paper enabled a comprehensive evaluation of the advantages and disadvantages of the two instructional delivery methods. We think this approach to be highly effective for online classes. Optimally, in online classes, students should have the freedom to select the class format that complements their own learning style, tailored to their environment and preferences.

This research proposes a function of SNS to reduce “stagnation of discussion, ” which hinders collaboration on SNS, as in the case of knowledge-building communities. Although knowledge-building may presuppose a kind of reciprocity where members perform their collective cognitive responsibilities, there are times when discussions cease and reciprocity appears to no longer be expected by participants, which may undermine their motivation to continue the collaboration. We have examined a method for delaying the appearance of posts on SNS by classifying them according to their time-sensitivity and the degree of author's contribution to the community. Such posts are displayed immediately after they are written, while ones with low time-sensitivity and posted by frequently contributing members are stocked for a while and displayed when the posting frequency decreases. This method can give the impression that the activity of the members was continuous, without stagnation of the discussion.

In Japan, data analysis has become an important part of the ‘Informatics’ subject in high schools. Data analysis is also becoming increasingly important at universities due to the promotion of ‘Mathematical, Data Science, and AI Education Programs’ and other programs. Students learn how to manipulate statistical processing and graph drawing using the computer. We therefore developed a system called ‘Connect DB’ that has an easy-to-learn operation system and supports learning data analysis. This system supports learners in learning data analysis, such as collecting, organizing, and formatting data, by enabling them to analyze data with just a few mouse operations and by suggesting appropriate analysis methods based on the type of data. In addition, the system also provides sample data that can be used in classes to support teachers. This paper describes the design and implementation of the Connect DB data analysis learning system. By comparing the number of operations with spreadsheet software and analyzing the post-training questionnaire for university students and their operation logs, we confirmed that this system is easy to learn and that it can be used for practical training in data analysis.

In the academic years (AY) 2020 and 2021, due to the COVID-19 pandemic, many lectures in universities had to be conducted online. Our faculty provided programming classes in-person instead of online because the programming exercise plays an essential role in classes. Moreover, face-to-face interactions in programming exercises are significant for instructors as they can intensively supervise students' work. The author created lecture videos for lessons provided as hybrid online courses in the AY 2020. Therefore, it was easy to move from regular classes to flipped classes utilizing such resources. In addition, the migration from in-person to online classes has resulted in the intensive use of a learning management system. It has also helped us introduce several learning analytics methods in practical educational fields. This paper provides an overview of the flipped classes for programming courses for first-grade university students conducted by the author in the AY 2021. It also discusses the relationship between students' attitudes toward participation and their performance.

The objective of this paper is to examine the consequences of Totsuka's practice. In the '80s, a terrific teacher developed his original version of Japanese LOGO by himself and tried it in a small elementary school in a rural area as “Totsuka's Practice”. He used LOGO programming as a tool for the empowerment of children's scientific research and help with mathematical understandings. We have conducted a longitudinal cohort study over 30 years of Totsuka's practice. The longitudinal cohort study suggested that 91 people out of 250 who were taught by Totsuka showed a different attitude toward university enrollment and advanced professions of their work from the average person. Further, an interview study of two people suggested that they did not care about making mistakes because there were many routes to reach a goal, and they enjoyed construction using scientific thinking and programming as a tool. It suggested that Totsuka's practice is still living in their minds, and computers might work as an amplifier for empowering themselves tacitly.

The purpose of introductory data science education for undergraduates is to increase students' interest in data science and motivate them to learn. In Japanese university students' standard data science curriculum, group work based on AI application examples is recommended to motivate students to learn data science. All university students who have just graduated from high school may not have sufficient basic data science knowledge, and their understanding of its importance may differ from person to person. Under these circumstances, it is unclear whether an introduction to AI applications would be practical as an introductory education to data science. Therefore, in this study, we designed a lesson including group work based on a case study of using AI technology in job interviews. Although first-year university students' willingness to learn AI differed by department, we found that their motivation in learning AI increased after the lesson.

This paper proposes a new method for control-point based geometric correction of historical maps. The proposed method is a hybrid approach combining the triangulated irregular network (TIN) method with the vector field analysis (VFA) method. The above new method aims to achieve a short calculation time that is an advantage of TIN while also maintaining both the homeomorphism property and straightness of designated specific edges that are advantages of VFA. Specifically, the control-point triangle mesh obtained by the TIN method is modified to avoid the violation of homeomorphism by adding new control points whose positions are estimated by the VFA method. The triangle mesh thus obtained is then further modified by the constrained-triangulation method so that the designated edges necessarily exist. The experimental results show that the proposed method actually achieves the above characteristics.

Hand gestures are communication signals that emphasize an important part of an utterance and express the concept of emphasized words. Iconic gestures are hand gestures that depict concrete actions, objects, or events mentioned in speech. In this study, assuming that gesture forms of iconic gestures are determined based on the image of a given object in the speaker's mind, we propose a method for selecting iconic gesture forms based on the image representation obtained from a set of pictures of an object. First, we asked annotators to select a gesture form that best expresses the meaning of a given word based on a typical image and concept in their minds. We also collected a set of pictures of each entity from the web and created an average image representation from them. We then created a Deep Neural Network (DNN) model that takes a set of pictures of objects as input and predicts the typical gesture form that originates from the human mind. In the model evaluation experiment, our two-step gesture form selection method successfully classified seven types of gesture forms with an accuracy of over 62%. Furthermore, we created character animations that performed selected gestures and conducted a preliminary perception study to examine how human users perceive animated iconic gestures.

The eBPF (Berkeley Packet Filter) in the Linux OS is a virtual machine for injecting user-space programs written in C language inside the Linux kernel, to perform a range of network processing functions, by attaching them to kernel level hooks such as system calls. Despite being a revolutionary replacement to in-kernel programming and being increasingly adopted by kernel-focused applications, developers struggle to understand and use eBPF directly. This is due to the conditions that for C programs to be able to run inside the Linux kernel, they need to be Non-Turing complete and successfully pass the safety checks of the eBPF verifier module inside eBPF. As C is a Turing-complete language, this puts the onus on developers to design a C program with restricted, Non-Turing complete functionality. In order to reduce the burden on developers, a Domain Specific Language called PRSafe was created. In this paper, we aim to provide an introduction to PRSafe and provide a qualitative evaluation between programs written for eBPF with conventional development toolchains vs PRSafe. We go further to use PRSafe in conjunction with K2, a synthesis compiler used for optimization and verification of eBPF code.

A maneuver coordination service (MCS) is a generic cooperative driving method for connected and automated vehicles. In an MCS, each vehicle periodically broadcasts its planned trajectory by vehicular communication. When a vehicle wants to change its planned trajectory, it can send a desired trajectory and initiate negotiation with surrounding vehicles. A vehicle that receives the desired trajectory decides whether to accept the negotiation and updates its planned trajectory so as not to conflict with the desired trajectory for coordination. Most previous works evaluated MCSs assuming that the number of vehicles is small and that negotiation is always accepted. However, in a congested environment, accepting all negotiations may not be optimal for traffic efficiency because accepting negotiation affects both the negotiation participants and the surrounding vehicles that are not involved in the negotiation. In this paper, we propose a negotiation acceptance scheme in maneuver coordination for efficient traffic. This paper also proposes a utility function that takes the negotiation effect into account. In addition, the agreement phase is incorporated into the negotiation protocol to maintain acceptance consistency among several negotiation participants. Simulation results show that making appropriate settings for the negotiation acceptance scheme improves traffic efficiency in congested environments.

Although machine learning methods with byte n-grams have been marking high score for classifying malware and benignware, they seem not to be used for current anti-virus software. A performance bottleneck of the methods is dealing with byte n-grams in preprocessing such as top-k selection. It takes a long time to extract all byte n-grams which are required for selecting top-k n-grams. Moreover, if several “n”s are wanted to be used such as 4-grams, 8-grams and 16-grams, n-grams with each “n” must be extracted again and again. Therefore, we proposed a fast preprocessing method of extracting n-grams by applying a suffix array algorithm. Furthermore, our method can manage multi-length byte n-grams at the same time. In addition, selecting feature n-grams like top-k n-grams with information gain is also included in our method. On the other hand, our method has a limitation that it is only applicable to a large number of samples in the same malware subspecies family, which become extinct. We evaluated the speed of our method by comparing with usual ways. We also evaluated our method by machine learning with actual samples in four old malware subspecies families. We think there is a hope that our method may be applicable to detecting current targeted malware.

A previous study on target pointing has shown that the accuracy of performance models improves as the number of participants and clicks increases, but the task was limited to artificially simplified one-dimensional movements. Practical user interfaces often require more complex operations, and thus we examine the effects of the number of participants and task repetitions on the fit of existing models for path-steering tasks. Empirical results showed that the model for predicting movement times consistently fitted the data with high accuracy, even when the numbers of participants and repetitions were small. However, the model for predicting error rates was less accurate in terms of R², MAE, and RMSE. Therefore, the benefit of recruiting numerous participants is relatively greater for the error-rate prediction model, which supports the previous study on target-pointing tasks.

In topic modeling, phrases act as important grammatical units that help users interpret the semantics of extracted topics. Embedding-based topic modeling, which has been proposed recently, is a promising approach to extracting phrase-level topics because it does not suffer from scalability issues due to the increased vocabulary size by adding phrases. However, the quality of the phrase-level topics extracted by this approach has not been evaluated, and the effect of the choice of the embedding models used for this method has not been investigated. In this paper, we validate the performance of the phrase-level embedding-based topic modeling and evaluate the effect of the embedding models on the quality of the phrase-level topics. From the result of the evaluation, we realized that the existing pre-trained BERT models have limitations in either sentence or phrase representation; therefore, we further propose a joint fine-tuning of BERT for phrase and sentence embeddings to improve the quality of phrase-level topic modeling. The experimental results quantitatively and qualitatively demonstrate that the jointly fine-tuned BERT yields more coherent phrase-level topics compared with other methods, including popular LDA-based phrase topic modeling.

Subtitles and closed captions, originally intended for individuals with hearing impairments, have gained widespread usage among non-hearing-impaired individuals. This study focuses on adapting subtitles and captions, particularly the incorporation of the kanji ruby, for non-hearing-impaired users. Our study investigated the necessity of kanji ruby by subjecting non-hearing-impaired adults to audio content through experiments employing Welch's t-test. Additionally, we proposed and evaluated an adaptive model for determining whether ruby should be added to kanji captions based on the experimental outcomes. The results indicate that optimizing kanji ruby requires considering not only the difficulty of the kanji characters and the user's proficiency but also the audio content.

Reducing power consumption is one of the most important issues in smartphones, especially for the CPU, since it is one of the most power-consuming devices. Improving the user experience by increasing CPU processing performance are also important. There is a trade-off between reducing CPU power consumption and improving the user experience. Decreasing the CPU clock rate reduces power consumption but degrades the user experience. Increasing the clock rate does the opposite. Therefore, it is desirable to increase the clock rate when and only when CPU resource consumption is high, and decrease it when it is low. However, the kernels of many smartphone operating systems, including the Linux kernel in the Android OS, use a follow-up policy of increasing or decreasing the clock rate after observing an increase or decrease in CPU resource consumption, which does not immediately provide appropriate clock rate control. We believe that predicting CPU usage in the near future will be critical for proper control. In this paper, we focus on the Android OS and propose a method to predict CPU usage in the near future by observing the behavior of foreground applications, and controlling the CPU clock rate based on the prediction. The proposed method modifies Android Runtime, which is the application execution environment, observes application method calls in Android Runtime, and predicts CPU usage in the near future based on these observations. We then demonstrate the effectiveness of the proposed method using our microbenchmark application and an actual distributed Android application.

An access control model called Zero Trust Architecture (ZTA) has attracted attention. ZTA uses the information of users and devices, called context, to verify access requests. Zero Trust Federation (ZTF) has been proposed as a framework for extending an idea of identity federation to support ZTA by sharing contexts among participants in the federation. ZTF defines Context Attribute Provider (CAP) as the entity that collects context and provides it to each organization (Relying Party; RP) that needs context for verification based on ZTA. For precise verification, CAPs need to collect context from various data sources. However, ZTF did not provide a method for collecting context from data sources other than RP although collecting contexts is essential to realize ZTF. In this research, as a general method for collecting context in ZTF, we propose a method of linking identifiers between the data source and CAP. Then, we implemented our method using RADIUS and MDM as data sources and confirmed that their contexts could be collected and used.

Machine learning is often used for malicious website detection, but an approach incorporating WebAssembly as a feature has not been explored due to a limited number of samples, to the best of our knowledge. In this paper, we propose JABBERWOCK (JAvascript-Based Binary EncodeR by WebAssembly Optimization paCKer), a tool to generate WebAssembly datasets in a pseudo fashion via JavaScript. Loosely speaking, JABBERWOCK automatically gathers JavaScript code in the real world, converts them into WebAssembly, and then outputs vectors of the WebAssembly as samples for malicious website detection. We experimentally evaluate JABBERWOCK from three perspectives. First, we measure its processing time. Second, we compare the samples generated by JABBERWOCK with the actual WebAssembly gathered from the Internet. Third, we investigate if JABBERWOCK can be used in malicious website detection. Regarding the processing time, we show that JABBERWOCK can construct a dataset in 4.5 seconds per sample for any number of samples. Next, comparing 10,000 samples output by JABBERWOCK with 168 gathered WebAssembly samples, we believe that the generated samples by JABBERWOCK are similar to those in the real world. We then show that JABBERWOCK can provide malicious website detection with 99% F1-score because JABBERWOCK makes a gap between benign and malicious samples as the reason for the above high score. We also confirm that JABBERWOCK can be combined with an existing malicious website detection tool to improve F1-scores. JABBERWOCK is publicly available via GitHub (https://github.com/c-chocolate/Jabberwock).

Community improvement is about enhancing the physical infrastructure and promoting social, economic, and health outcomes. For comprehensive community enhancement, this paper proposed an analysis model to identify multiple clusters using the Cartesian product of network motifs and local-determined keywords. These clusters have the potential to intersect various domains and disciplines, fostering a more holistic understanding of community phenomena. Furthermore, we demonstrate the practical application of the model through two case studies: tourism and second-hand luxury goods transactions. Our findings in case studies reveal the potential of network motifs in identifying clusters that have the possibility of contributing to community improvement to some degree. These results have significant potential implications for both theoretical research and practical applications in community improvement, providing a new approach to identifying multiple clusters across diverse activities.

RPL is a standard routing protocol for an IPv6-Based LLN (Low power and Lossy Network). In RPL, each node estimates link quality to neighboring nodes and selects its parent node by an OF (Objective Function) to construct a tree structure rooted at the sink node. Currently, a typical OF is MRHOF (Minimum Rank with Hysteresis OF). MRHOF on Contiki-NG, an operating system for LLN devices, adopts an active monitoring approach for link quality estimation, which consumes much battery power. This paper proposes BROF (Broadcast Reception based OF) based on a passive monitoring approach. BROF listens to control messages transmitted by neighboring nodes and exploits unicast messages transmitted to the parent node. The performance of BROF and MRHOF on Contiki-NG is evaluated with a simulator and a testbed. The evaluation results show that the delivery ratio and delay of data messages in both OFs are almost the same while the number of transmitted L2 frames for control messages in BROF is only 19.7% to 37.8% of that in MRHOF on Contiki-NG for constructing and maintaining a tree structure. This means that BROF can achieve almost the same performance for sensing data collection as MRHOF on Contiki-NG with remarkably less battery consumption.

Generally, second or foreign language (SFL) classes are unable to provide sufficient opportunities for personalized speaking practice due to the time constraints of classroom hours. For decades, existing computer-mediated feedback (CMF) studies have proved that out-of-class speaking practice with audio recording effectively improves learners' speaking performance. Although such CMF studies have enabled language teachers to provide teacher electronic feedback (TEF) outside classroom hours, the best method of providing TEF had not been established and the effect of TEF had been unclear in speaking classes. This research proposes a practical method of providing TEF using an online oral repetition practice support system, ‘ORP Gym’. To investigate the effect of TEF, the controlled experiment was conducted on an online Japanese SFL basic course at an Indian university. Six professional online evaluators were trained and employed to generate TEF. Sixty-seven Indian undergraduate students, all are beginners, participated in the experiment. As the result of the pretest and posttest analysis, the experimental group who practiced with TEF showed statistically significantly greater learning gain in grammar error categories than the control group who practiced without TEF. Since ORP Gym can be deployed for other language courses, the proposed method promotes the further investigation of TEF in different levels and language courses.

In various problems on graphs to find the optimal path or circuit, feasible solutions are normally assumed to be simple. In reality, however, one-way or return roads, bridges connecting enclaves, etc. often lead to multiple passes through the same points or roads, i.e. a non-simple path or circuit. The difficulty in considering a non-simple path or circuit is that conventional subtour elimination constraints are of no use. Taking the orienteering problem as a target, we newly develop a subtour elimination constraint that works directly against non-simple type problems, and show its effectiveness in computational experiments.

The symmetric-key block cipher SLIM, which was proposed by Aboushosha et al., is a lightweight cryptographic algorithm. Designed for radio frequency identification (RFID) systems, SLIM is a 32-bit block cipher based on the Feistel structure with an 80-bit secret key. We present a higher-order differential attack on reduced-round SLIM. We discovered some 9-round higher-order differential characteristics from computer experimentation. We show that 12-round SLIM is attackable with 2³¹ data and 2^77.1 times of encryption using the 31st-order differential characteristic. Because the recommended number of rounds is 32, SLIM is demonstrated as secure against higher-order differential attack.

Interactive machine learning (IML) allows users to build their custom machine learning models without expert knowledge. While most existing IML systems are designed with classification algorithms, they sometimes oversimplify the capabilities of machine learning algorithms and restrict the user's task definition. On the other hand, as recent large-scale language models have shown, natural language representation has the potential to enable more flexible and generic task descriptions. Models that take images as input and output text have the potential to represent a variety of tasks by providing appropriate text labels for training. However, the effect of introducing text labels to IML system design has never been investigated. In this work, we aim to investigate the difference between image-to-text translation and image classification for IML systems. Using our prototype systems, we conducted a comparative user study with non-expert users, where participants solved various tasks. Our results demonstrate the underlying difficulty for users in properly defining image recognition tasks while highlighting the potential and challenges of interactive image-to-text translation systems.

Cascadic Parareal is a parallel-in-time (PinT) method developed to improve the parallel performance for explicit time-dependent problems. It is more efficient than other PinT methods when explicit methods are used for solving. Cascadic Parareal has been proven to accelerate a one-dimensional advection problem and a two-dimensional compressible flow simulation faster compared to spatial parallelism with more than 64 cores in the previous works. However, Cascadic Parareal has also demonstrated slow convergence and produced unstable results for supersonic flow simulations. The instability is caused by unstable supersonic flow results calculated on the coarse meshes. In the present work, we introduce an improvement for Cascadic Parareal using local mesh refinement (LMR) to improve its accuracy for supersonic flow simulations. Numerical experiments in this research demonstrate that the LMR method can improve the convergence rate and accuracy of Cascadic Parareal for supersonic flow simulations. The numerical experiments of the present work show that the improved PinT method can provide stable and more accurate simulations for supersonic flow, and the compute time performance of the PinT algorithm can outperform simple spatial parallelism.

Kotlin is a newly proposed programming language that is highly compatible with Java. Typically, source code written in Kotlin is compiled into Java bytecode, which is then interpreted and executed by the Java Virtual Machines (JVMs). To achieve high performance, the codes (source code or bytecode) must be optimized for the existing Java virtual machines. In this paper, we evaluate the performance of loop processing in Kotlin and Java programs considering the existing JVM implementations. First, we perform micro-benchmarking of loop processing, such as for and while statements, using two popular JVM implementations. The results show that the performance depends on the description methods even for the same semantics. The performances can be classified into two groups: fast and slow, in both JVM implementations. Second, we compare the bytecodes generated by the compilers from the description methods of the fast and slow groups. We then show the differences between them, which are small and cannot be justified as the direct cause of a significant performance difference. Third, we compare the native codes generated by a just-in-time (JIT) compiler and show that the bytecode in the fast group is deeply optimized by the JIT compiler, while that in the slow group is not. In fact, small differences in the bytecode lead to differences in the behavior of the JIT compiler and to non-trivial performance gains.