記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

In recent years, Siamese network-based trackers brought new vitality to the visual object tracking field. However, tracking tasks have always been troubled by complex scenarios. As Siamese trackers become more powerful, the performance bottlenecks caused by complex scenarios become more and more non-negligible. Occlusion is the most common and challenging complex scenario that can easily cause tracking failures. Some high-quality tracking databases provide visible ratio labels to describe occlusion in more detail. In addition, high-performance Siamese trackers can not run efficiently on resource-limited devices due to their high memory cost and complexity. To address these issues, we propose an Adaptive Multi-teacher Knowledge Distillation (AMKD) model to distill lightweight tracker, which is fast and achieves satisfactory performance in low visible ratios scenarios. In AMKD, we adopt the teacher model to transfer adequate knowledge to student. Furthermore, to extract visibility-based knowledge from visible ratios labeled data and transfer it to student efficiently, we introduced assistant teachers which are customed to overcome low visible ratios scenarios. For multiple assistant teachers transfer knowledge to student more efficiently and effectively, the AMKD is equipped with an Adaptive Selection Mechanism (ASM). Experiments of several Siamese trackers on high-quality dataset GOT-10K demonstrated the effectiveness of our method. Moreover, the AMKD distilled student achieve 9 times of compression rates and 6 times of speed up reach 181 FPS while improving accuracy in low visible ratios scenarios and obtaining favorable overall performance.

DOI： 10.1109/ISAS61044.2024.10552525

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85197636958&origin=inward

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

In using large language models (LLMs) for education, such as distractors in multiple-choice questions and learning by teaching, error-containing content is used. Prompt tuning and retraining LLMs are possible ways of having LLMs generate error-containing sentences in the learning content. However, there needs to be more discussion on how to tune LLMs for specific lecture content. Such discussions help control LLMs and for developing educational applications. In this study, we aim to train a detuned LLM that only states incorrect things, considering the limitations of prompt-based approaches such as prompt injection. Our method detunes LLMs by generating datasets that confuse LLMs. To evaluate our method, we asked the detuned LLM to solve multiple-choice questions to evaluate whether it answered the questions incorrectly or not. We also evaluate how many errors are contained in the sentences generated by the LLM to investigate how their knowledge of lecture content is degraded regarding factuality.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85214024496&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Learning logs collected by digital educational systems, increasingly deployed in educational settings, include clickstream logs recorded through page transitions in teaching materials and digital marker logs recorded by drawing a marker. A challenge with these learning logs is their low temporal and spatial resolutions. This paper proposes a system that generates a high-resolution learning log (HLL) by utilizing learners' gaze information obtained through webcam-based eye tracking. We also propose methods for analyzing learners' learning-theme browsing patterns using HLL. The HLL retains the attention time of the learning-themes on the learning material and viewing time in and out of the screen. Utilizing the HLL allows learners' attention transitions to be captured over time. Compared with traditional topic-based learning log analysis methods, HLL offers a more granular analysis of detailed learning theme browsing patterns.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85213982105&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Course planning is essential for academic success and the achievement of personal goals. Although universities provide course syllabi and curriculum maps for course planning, integrating and understanding these resources by the learners themselves for effective course planning is time-consuming and difficult. To address this issue, this study proposes a method that uses generative AI to classify relationships between subjects and generate explanatory texts describing the connections of subjects and positions of subjects within the curriculum based on subject and curriculum information. An evaluation experiment involving learners demonstrated a classification accuracy of approximately 70% for inter-subject relationships. Furthermore, our experimental results confirm that that the generated explanatory texts significantly enhance the understanding of relationships between subjects, and are thus effective for course planning.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85214023587&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

This study explores the effectiveness of Large Language Models (LLMs) for Automatic Question Generation in educational settings. Three LLMs are compared in their ability to create questions from university slide text without fine-tuning. Questions were obtained in a two-step pipeline: first, answer phrases were extracted from slides using Llama 2-Chat 13B; then, the three models generated questions for each answer. To analyze whether the questions would be suitable in educational applications for students, a survey was conducted with 46 students who evaluated a total of 246 questions across five metrics: clarity, relevance, difficulty, slide relation, and question-answer alignment. Results indicate that GPT-3.5 and Llama 2-Chat 13B outperform Flan T5 XXL by a small margin, particularly in terms of clarity and question-answer alignment. GPT-3.5 especially excels at tailoring questions to match the input answers. The contribution of this research is the analysis of the capacity of LLMs for Automatic Question Generation in education.

DOI： 10.1007/978-3-031-72312-4_18

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85205312201&origin=inward

記述言語：英語   掲載種別：研究論文（学術雑誌）

Student performance prediction was deployed to predict learning performance to identify at-risk students and provide interventions for them. However, prediction models should also consider external factors along with learning activities, such as course duration. Thus, we aim to distinguish the difference factor between the time dimension (duration of the course) and the feature dimension (students' learning activities) by attention weights to provide helpful information and improve predictions of student performance. In this study, we introduce Attention-Based Artificial Neural Network (Attn-ANN), a novel model in educational data mining. The Attn-ANN combines attention weighting on the time and feature dimensions to examine the significance of lectures and learning activities and makes predictions by visualizing attention weight. We found that the Attn-ANN had a better area under the curve scores than conventional algorithms, and the attention mechanism allowed models to focus on input selectively. Incorporating the attention weighting of both the time and feature dimensions improved the prediction performance in an ablation study. Finally, we investigated and analyzed the model's decision, finding that the Attn-ANN may be able to create synergy in real-world scenarios between the Attn-ANN's predictions and instructors' expertise, which underscores a novel contribution to engineering applications for interventions for at-risk students.

DOI： 10.1109/ACCESS.2024.3429554

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85199673606&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

In recent years, with the rapid development of neural networks, visual object tracking is becoming increasingly important in real-world applications such as camera drones and driving assistant systems, even self-driving technology. Neural network-based Siamese trackers achieve satisfactory accuracy in the object tracking field and stand out. However, high-performance Siamese trackers are often designed to be complex and heavy, which hinders their application on resource-limited mobile devices. Thus, compressing the neural network-based tracker to make it lightweight and efficient without obvious performance degradation is of great significance. Inspired by the outstanding work of RepVGG which focuses on compressing multi-branches neural networks without any accuracy cost, we propose the RepSiamses Tracker (RST) which is extremely lightweight and achieves very high tracking speed. In RST, the RepVGG-based backbone depth can be adapted to the different target hardware which benefits from the high flexibility of RepVGG. The experimental results on the high-quality benchmark VOT2018 and LaSOT show that RST achieves satisfactory accuracy and extremely high tracking speed on GPU. More impressively, RST is able to run on the CPU at a hyper-real-time of 69 fps and with very little memory cost of 16.4 MB. Such high tracking speed and low memory cost can bridge the gap between academic algorithms and real-world applications.

DOI： 10.1109/ISAS61044.2024.10552459

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85197588562&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Understanding students' coding behaviors is crucial for providing targeted support in programming education. Automatic analysis of coding behaviors using machines can address the limitations of manual monitoring. Previous studies focused on coding behavior representations without considering differences relative to a model answer. We propose embedding vectors that capture these differences, enabling the distinction between simple and complex code solutions. Evaluating these vectors by predicting assignment scores, we achieved over 15% higher accuracy compared to conventional methods. This approach has the potential to enhance teachers' understanding of students' coding behaviors and improve support in programming education.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85181532751&origin=inward

記述言語：英語   掲載種別：研究論文（学術雑誌）

This study helps improve the early prediction of student performance by RNN-FitNets, which applies knowledge distillation (KD) to the time series direction of the recurrent neural network (RNN) model. The RNN-FitNets replaces the teacher model in KD with “an RNN model with a long-term time-series in which the features during the entire course are inputted” and the student model in KD with “an RNN model with a short-term time-series in which only the features during the early stages are inputted.” As a result, the RNN model in the early stage was trained to output the same results as the more accurate RNN model in the later stages. The experiment compared RNN-FitNets with a normal RNN model on a dataset of 296 university students in total. The results showed that RNN-FitNets can improve early prediction. Moreover, the SHAP value was employed to explain the contribution of the input features to the prediction results by RNN-FitNets. It was shown that RNN-FitNets can consider the future effects of the input features from the early stages of the course.

DOI： 10.1177/07356331221129765

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85141013440&origin=inward

記述言語：英語   掲載種別：研究論文（学術雑誌）

In this study, we propose an integrated system to support learners' reviews. In the proposed system, the review dashboard is used to recommend review contents that are adaptive to the individual learner's level of understanding and to present other information that is useful for review. The pages of the digital learning materials that are estimated to be insufficiently understood by each learner and the webpages related to those pages are recommended. As a method for estimating such pages, we consider extracting the pages related to the questions that were answered incorrectly. We examined the accuracy of matching each question with the pages of the learning materials. We also conducted an experiment to verify the usefulness of the system and its effect on learning using a review dashboard. In the experiment, the evaluation of the review dashboard indicated that at least half of the participants found it useful for most types of feedback. In addition, the rate of change in quiz scores was significantly higher in the group using the review dashboard, which indicates that using the review dashboard has the effect of improving learning.

DOI： 10.1109/TLT.2022.3225206

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85144038153&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Recent digital material delivery systems enable teachers not only to upload lecture materials but to analyze students’ behavior, such as browsing data with detailed operation logs that record which student performed which operation on which page at which time. While such behavioral data has been elucidated to be useful for predicting students’ performance in existing studies, it has yet to be fully verified how content (e.g., learning materials) information can be integrated with behavioral data. This paper proposes methods to utilize content information jointly with behavioral data and compares them with the baseline method using only behavioral data. The results indicate that one of the proposed methods performs better prediction of quiz-score prediction. This suggests that both the browsing behavior of students and the content information have an impact on student performance.

DOI： 10.1007/978-3-031-36336-8_86

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85164966449&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

The main objective of this paper is to improve the image segmentation model for handwritten notebook analytics. We conducted a considerable amount of research in this area to increase the accuracy and efficiency of segmentation. To address the issues with traditional methods, we introduced attention mechanism and recursive residual convolutional neural network in the multi-task U-Net model. Through training and testing the model on handwritten notebook dataset and compared it with other existing technologies, we demonstrated the effectiveness of this method. The results showed that the model had a significant improvement in accuracy. Therefore, the research findings in this paper are important for improving the technology of handwritten notebook analytics.

DOI： 10.1109/ICIP49359.2023.10222873

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85180799905&origin=inward

担当区分：責任著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Smart agriculture has assumed increasing importance due to the growing age of farmers and a shortage of farm leaders. In response, it is crucial to provide more opportunities to learn about smart agriculture at agricultural colleges and high schools, where new farmers are trained. In agricultural education, a system is used for managing environmental information, such as temperature and humidity, obtained from sensors installed in the field. However, it is difficult to make effective use of this system due to the time required to detect changes in the field interfering with class time and the problem of oversight. In this study, we proposed a field environment digest system that will help learners by providing the summarized field sensing information, and support them in analyzing the data. In addition, to examine the potential for using field sensing information in agricultural education, we investigated the usefulness of the summarized sensor information and students’ usage of this information. In this paper, we outline the contents of the developed system and the results of the digest evaluation experiments.

DOI： 10.1007/978-3-031-43393-1_10

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85174486166&origin=inward

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

When students use e-learning systems such as learning management systems and e-book systems, the operation logs are stored and analyzed to understand student learning behaviors. For implementing some applications, such as dashboard systems and at-risk student detection, the operation logs are mainly transformed into features designed by researchers. Such hand-crafted features, like the number of operations, are easily interpretable. However, the power of the hand-craft features may be limited for the recent large-scale educational dataset. In machine learning research, data-driven features are demonstrated to be a better representation than hand-crafted features. However, there are few discussions in the educational data due to a need for many operation logs. In this study, we collect reading logs of an e-book system. We propose a representation learning method for the reading logs based on contrastive learning. Our proposed method transforms time-series reading logs into reading behavior feature vectors directly without hand-crafted features. In our experiments, we demonstrate that the power of our feature representation is better than a traditional count-based hand-crafted feature representation in the at-risk student detection task. In addition, we investigate the characteristics of the feature space learned by our proposed method.

DOI： 10.1007/978-3-031-36272-9_35

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85164920785&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Teaching and learning from one another is one of the most effective ways for learners to acquire proactive learning attitudes. In this study, we propose a new learning support system that encourages mutual teaching and learning by introducing a mechanism that guarantees sustainability. Learners submit articles called “learning-articles” that summarize their own learning and knowledge. The proposed system not only accumulates and publishes these articles but also has a mechanism to encourage the submission of necessary topics. The proposed system has been in operation since the academic year 2020, and it has collected learning-articles across our university’s nine academic disciplines from more than 300 learners. To investigate the effects of sharing learning-articles on education from the learners’ perspectives, a questionnaire was distributed among 25 students.

DOI： 10.1007/978-3-031-43393-1_44

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85174499041&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

The use of e-book reading systems such as Bookroll, and their ability to record readers' activities allows the design of predictive models capable of identifying at-risk students from their reading characteristics. Even though previous works have obtained promising results in this task, these results may not evidence the expected prediction performance in practical applications due to their selected assessment methods. Accordingly, in this paper, we assess this performance in two practical scenarios. The first is when we keep stored data from previous years of our course which can be used to train our model, and the second is when we only have data from a different course to use in this training process. In order to obtain a more accurate assessment, we collected 92, 574 samples of predictive performances from different models under the above-mentioned conditions. We also considered different feature representations along with variational latent representations, which can leverage our previous data to automatically design general hidden features. From our results, we understand that in the first condition we can expect a relatively good predictive performance, especially when using variational latent representations. However, in the second condition we found that even when using them, the predictive performances are very limited resulting in an impractical solution.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85151045820&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

In recent years, several kinds of e-learning systems, such as e-book and Learning Management System (LMS) have been widely used in the field of education. When students access these systems, their activities on the systems will be continuously and automatically recorded and stored as learning logs. As the learning logs are stored in association with students and indicate students' learning activities, most studies have been “student-based” learning log analyses focused on students and each student's learning behavior. However, the “student-based” learning log analysis focuses on each student's learning behavior during the entire lesson (for example, studied well or didn't study enough) and cannot show what they learned. Therefore, if there is a need to investigate students' learning behavior regarding each topic of the lesson, such as which topic is learned well and which not in order to optimize the syllabus, we cannot conduct “student-based” learning log analysis directly. Instead of “student-based” learning log analyses, this study describes a method of “learning-topic-based” learning log analysis. We will show how to convert a learning log associated with students into a learning-topic-associated one and shape the logs into a two-dimensional matrix of learning topics and learning activities. Then we apply Non-negative Matrix Factorization (NMF) to the matrix in order to extract the learning patterns by activity. In addition, we make a three-dimensional matrix (tensor) of students, learning topics, and learning activities by subdividing the learning activities of each learning topic by students. We then apply Non-negative Tensor Factorization (NTF) to the tensor to extract detailed learning patterns. The methods proposed in this study will help teachers to have a comprehensively view of students' learning behaviors towards each learning topic easily even if the learning log is in a large-scale, so teachers can adjust syllabus according to the attracted learning behaviors, which is helpful to increase learning efficiency.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85151062844&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Since the demand for programmers is increasing, programming courses are being offered widely. In this context, students' motivation can be damaged by difficulties they encounter in their programming courses. Although teachers' support is necessary to prevent such an issue, it is impossible for teachers to directly monitor all students' programming activities at the same time and determine which students have troubles with programming. Therefore, several studies have been conducted to help teachers monitor students. However, these studies do not provide an understanding of the activities of students who do not run their code, which may lead researchers to miss students who are in trouble. In this paper, we propose an indicator for detecting students who need coding support by analyzing programming logs that are recorded even when the students do not run their code. This gives teachers deeper insight into the students' programming performance. Although further work remains, the validation of this indicator shows that it could detect those students who are in trouble.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85151055172&origin=inward

記述言語：英語   掲載種別：研究論文（学術雑誌）

Automatic detection and counting of vehicles in a video is a challenging task and has become a key application area of traffic monitoring and management. In this paper, an efficient real-time approach for the detection and counting of moving vehicles is presented based on YOLOv2 and features point motion analysis. The work is based on synchronous vehicle features detection and tracking to achieve accurate counting results. The proposed strategy works in two phases; the first one is vehicle detection and the second is the counting of moving vehicles. Different convolutional neural networks including pixel by pixel classification networks and regression networks are investigated to improve the detection and counting decisions. For initial object detection, we have utilized state-of-the-art faster deep learning object detection algorithm YOLOv2 before refining them using K-means clustering and KLT tracker. Then an efficient approach is introduced using temporal information of the detection and tracking feature points between the framesets to assign each vehicle label with their corresponding trajectories and truly counted it. Experimental results on twelve challenging videos have shown that the proposed scheme generally outperforms state-of-the-art strategies. Moreover, the proposed approach using YOLOv2 increases the average time performance for the twelve tested sequences by 93.4% and 98.9% from 1.24 frames per second achieved using Faster Region-based Convolutional Neural Network (F R-CNN) and 0.19 frames per second achieved using the background subtraction based CNN approach (BS-CNN), respectively to 18.7 frames per second.

DOI： 10.1007/s11042-022-12370-9

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85127200112&origin=inward

記述言語：英語   掲載種別：研究論文（学術雑誌）

The COVID-19 pandemic has led to an increase in online classes, and programming classes are no exception. In such a learning environment, understanding every student’s programming process is mostly impractical for teachers, despite its significance in supporting students. Giving teachers feedback on programming processes is a typical approach to the problem. However, few studies have focused on visual representations of the evolution process of source-code contents; it remains unclear what visual representation would be effective to this end and how teachers value such feedback. We propose two feedback tools for teachers. These tools visualize the temporal evolution of source-code contents at different granularities. An experiment was conducted in which several university teachers performed a user evaluation of the tools, particularly with regard to their usefulness for reviewing past programming classes taught by another teacher. Questionnaire results showed that these tools are helpful for understanding programming processes. The tools were also found to be complementary, with different aspects being highly evaluated. We successfully presented concrete visual representations of programming processes as well as their relative strengths and weaknesses for reviewing classes; this contribution may serve as a basis for future real-time use of these tools in class.

DOI： 10.3390/su14138084

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85133658987&origin=inward

記述言語：英語   掲載種別：研究論文（学術雑誌）

While positive effects of imitating other learners have been reported, the recent increases in the number of online classes have seriously limited opportunities to learn how others are learning. Providing information about others’ learning activities through dashboards could be a solution, but few studies have targeted learning activities on e-textbook systems; it remains unclear what information representations would be useful and how they would affect learning. We developed a dashboard system that enables live sharing of students’ learning activities on e-textbooks. An experiment was conducted applying the dashboard in an online class to evaluate its impact. The results of questionnaires and quizzes were analyzed along with learning activities on the e-textbook system. From the questionnaire results, the most useful feedback types were identified. Regarding the impact on learning, the study found that a higher percentage of students who used the dashboard followed the progress of the class than those who did not. The study also found that students who used the dashboard were more likely to achieve higher quiz scores than those who did not. This study is the first to reveal what specific feedback is useful and to successfully investigate the impact of its use on learning.

DOI： 10.3390/su14126946

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85132025063&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Handwritten notes are one important component of students' learning process, which is used to record what they have learned in class or tease out knowledge after class for reflection and further strengthen the learning effect. It also helps a lot during review. We hope to divide handwritten notes (Japanese) into different parts, such as text, mathematical expressions, charts, etc., and quantify them to evaluate the condition of the notes and compare them among students. At the same time, data on students' learning behaviors in the course are collected through the online education platform, such as the use time of textbook and attendance, as well as the scores of the online quiz and course grade. In this paper, the analysis of the relationship between the segmentation results of handwritten notes and learning behavior are reported, as well as the research on automatic page segmentation based on deep learning.

DOI： 10.1145/3506860.3506915

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85126210107&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Students' performance prediction is essential for instructors to observe each student's learning behavior to discover which students have become at-risk. Early prediction helps instructors to intervene in time and provide academic support to these students. However, instructors should grasp essential behavior points to survey students' academic performance. In this study, we propose the Scaled-Dot Product Attention that can mine the relationship between the student's learning behaviors and performance to find the essential features that directly affect students' performance. In this study, we tested the early prediction performance of Scaled-Dot Product Attention with conventional algorithms. We then investigated essential lectures or features related to students' learning activities. From the result, we found that Scaled-Dot Product Attention outperformed the conventional algorithms to identify at-risk students and found the important lectures and students' actions.

DOI： 10.1109/TALE54877.2022.00059

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85163832702&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Student Performance Prediction is essential for learning analysis of the students' learning behavior to discovering at-risk students for the early invention to support students. This study transforms the students' reading behavior into a two-dimensional matrix input based on each lecture material's reading behavior. The matrix input will be updated by accumulating the value for each week for performance prediction week by week. The multilayer perceptron neural network is employed to receive the matrix input and give feedback as a student's criteria consist of at-risk or no-risk students. This study considers the accuracy of a model considering between on contents information and weekly information. We also investigate the switching of learning materials' order, the feature importance of the reading operation on an event stream, and the difference in reading behavior between at-risk and no-risk students. These can help the instructors for an early invention to support at-risk students.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85128884472&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

The emergence of digital textbook reading systems such as Bookroll, and their ability of recording reader interactions has opened the possibility of analyzing the students reading behaviors and characteristics. To date, several works have conducted compelling analyses characterizing the different types of students with the use of clustering ML models, while others have used supervised ML models to predict their academic performance. The main characteristic these models share is that internally they simplify the students' data into a latent representation to get an insight or make a prediction. Nevertheless, these representations are oversimplified, otherwise difficult to interpret. Accordingly, the present work explores the use of Variational Autoencoders to make more interpretable and complex latent representations. After a brief description of these models, we present and discuss the results of four explorative studies when using the LAK22 Data Challenge Workshop datasets. Our results show that the probabilistic latent representations generated by the proposed models preserve the student reading characteristics, allowing a better visual interpretation when using 3 dimensions. Also, they allow supervised regressive and classification models to have a more stable and less overfitted learning process, which also allows some of them to make better score predictions.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85128933513&origin=inward

担当区分：責任著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Background subtraction networks outperform traditional hand-craft background subtraction methods. The main advantage of background subtraction networks is their ability to automatically learn background features for training scenes. When applying the trained network to new target scenes, adapting the network to the new scenes is crucial. However, few studies have focused on reusing multiple trained models for new target scenes. Considering background changes have several categories, such as illumination changes, a model trained for each background scene can work effectively for the target scene similar to the training scene. In this study, we propose a method to ensemble the module networks trained for each background scene. Experimental results show that the proposed method is significantly more accurate compared with the conventional methods in the target scene by tuning with only a few frames.

DOI： 10.1109/AVSS56176.2022.9959316

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85143905004&origin=inward

担当区分：責任著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Automatic solar panel inspection systems are essential to maintain power generation efficiency and reduce the cost. Thermal images generated by thermographic cameras can be used for solar panel fault diagnosis because defective panels show abnormal temperature. However, it is difficult to identify an anomaly from a single panel image when similar temperature features appear in normal panels and abnormal panels. In this paper, we propose a different feature based method to identify defective solar panels in thermal images. To determine abnormal panel from input panel images, we apply a voting strategy by using the prediction results of subtraction network. In our experiments, we construct two datasets to evaluate our method: the clean panels dataset which is constructed by manually extracted panel images and the noise containing dataset which is consisting of panel images extracted by the automatic panel extraction method. Our method achieves more than 90% classification accuracy on both clean panels dataset and noise containing dataset.

DOI： 10.1117/12.2586911

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85112463605&origin=inward

担当区分：最終著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Recurrent neural network (RNN) achieves state-of-the-art in several researches of the performance prediction. However, accuracy in early time steps is lower than that in late time steps, even though the early detection of at-risk students is important for timely interventions. To improve the accuracy in early time steps, we propose a knowledge distillation method for RNN. Our method distills the time-series information in the RNN model of late time steps into the RNN model of early time steps. This distillation makes the prediction of early time steps closer to that of late time steps. The experimental result showed that our method improved the detection rate of at-risk students compared with traditional RNNs, especially in early time steps.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85175005202&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

The introduction of ICT technology into education has been attracting attention, and various educational platforms and tools have been developed. With the emergence of standards such as Learning Tools Interoperability, it has become possible to call external tools from Learning Management Systems in a unified manner. However, it is still challenging to integrate the tools and use them as a seamless learning environment. For this reason, people tend to develop a monolithic tool that integrates functions that meet individual needs. Repeatedly reimplementing similar functions leads to the mess of various learning log formats and data dispersion across the tools, making it hard to realize learning analytics across multiple learning support systems of different educational institutions. In order to solve this problem, we propose the concept of Compositional Learning Environments, with which we can combine any tools to form a new and more complex one. This demonstration shows two practical examples of combinations of an e-textbook tool with other types of learning support tools.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85109699560&origin=inward

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Detection of at-risk students is a fundamental issue in enhancing learning supports, and has been proposed based on students' learning activity in learning analytics. However, it is not clear which activity we should focus on to detect at-risk students such as low performance students. In this study, we proposed a clustering-based method for at-risk student detection based on three main clusters of students: inactive, passive, active students. Our method focused on reading behaviors and action behaviors in an e-book system. In addition, we consider which period of learning activities is effective for detecting at-risk students. The learning logs of 289 students of Cyber-Security course were collected for our analysis. In our comparison at different moment during the lecture, we found that the cluster of inactive students detected after 35 minutes of lecture got significant lower grades than other students, when the lecture was not too short nor too easy.

DOI： 10.1109/TALE48869.2020.9368453

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85102971043&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

In recent years, online classes have been increasingly conducted in various situations. However, in these classes, especially non-face-to-face and large-scale ones, it is more difficult for teachers and students to understand the status of the class during a lecture. To address this issue, we propose a real-time learning analytics dashboard that provides summarized information on teachers' instruction and students' learning activities during lectures. In this article, we introduce the real-time learning analytics dashboard and report its effectiveness through experiments in an online class at our university.

DOI： 10.1109/TALE48869.2020.9368340

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85102970196&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Recent enrichment of digital learning environments has made it possible to obtain learning logs (data) on learners' learning behavior. In this situation, it is possible to recommend learning contents which are appropriate for individual learner by analyzing learning data. Our study develops a learning support system which recommends personalized review materials based on the results of quizzes and learning activities recorded by e-textbooks. In this paper, we explain the details of the system and report experimental results.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85099578669&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

This paper introduces an open source library for e-Book (digital textbook) log analysis, called OpenLA. An e-Book system is a useful system which records learning logs. Various analysis using these logs have been conducted. Although there are many common processes in preprocessing logs, the functions have been developed by per researcher. To reduce such redundant development, OpenLA provides useful modules to load course information, to convert learning logs into a more sophisticated representation, to extract the required information, and to visualize the data. OpenLA is written in the Python language and compatible with other Python libraries for analysis. This paper provides a brief explanation of each module, followed by re-implementation samples of related studies using OpenLA. The details about OpenLA is open to public at https://www.leds.ait.kyushu-u.ac.jp/achievements.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85099475020&origin=inward

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Recently, deep neural networks have demonstrated excellent performance in foreground segmentation tasks such as moving object detection and change detection tasks. Various types of neural networks have been proposed, however, the previous works mainly discuss the accuracy. Analytics of the neural networks is important to utilize them effectively and improve their performance. In this paper, we investigate a foreground segmentation network and background subtraction network. In our analysis, we discuss differences of behaviors of the two networks in specific scenes and feature distributions in each layer of a background subtraction network to investigate feature learning. In addition, we provide suggestions about the comparison with these networks.

DOI： 10.1109/ICIP40778.2020.9191151

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85098649756&origin=inward

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Recently, various educational data such as clickstream data and eye movement data have been collected from students using e-learning systems. Learning analytics-based approaches also have been proposed such as student performance prediction and a monitoring system of student learning behaviors for supporting teachers. In this paper, we introduce our recent work as instances of the use of clickstream data and eye movement data. In our work, the clickstream data is used for representing student learning behaviors, and the eye movement data is used for estimating page areas where the student found difficulty. Besides, we discuss advantages and disadvantages depending on the types of educational data. To discuss them, we investigate a combination of highlights added on pages by students and eye movement data in page difficulty estimation. In the investigation, we evaluate the similarity between positions of highlights and page areas where the student found difficulty generated from eye movements. It is shown that areas in the difficult pages correspond to the highlights in this evaluation. Finally, we discuss how to combine the highlights and eye movement data.

DOI： 10.1007/978-3-030-50344-4_42

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85088749401&origin=inward

担当区分：責任著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Public datasets used to train modern object detection models do not contain all the object classes appearing in real-world surveillance scenes. Even if they appear, they might be vastly different. Therefore, object detectors implemented in the real world must accommodate unknown objects and adapt to the scene. We implemented a framework that combines background subtraction and unknown object detection to improve the pretrained detector’s performance and apply human intervention to review the detected objects to minimize the latent risk of introducing wrongly labeled samples to the training. The proposed system enhanced the original YOLOv3 object detector performance in almost all the metrics analyzed, and managed to incorporate new classes without losing previous training information.

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85083524629&origin=inward

記述言語：英語   掲載種別：研究論文（学術雑誌）

Automatic vehicle detection and counting are considered vital in improving traffic control and management. This work presents an effective algorithm for vehicle detection and counting in complex traffic scenes by combining both convolution neural network (CNN) and the optical flow feature tracking-based methods. In this algorithm, both the detection and tracking procedures have been linked together to get robust feature points that are updated regularly every fixed number of frames. The proposed algorithm detects moving vehicles based on a background subtraction method using CNN. Then, the vehicle’s robust features are refined and clustered by motion feature points analysis using a combined technique between KLT tracker and K-means clustering. Finally, an efficient strategy is presented using the detected and tracked points information to assign each vehicle label with its corresponding one in the vehicle’s trajectories and truly counted it. The proposed method is evaluated on videos representing challenging environments, and the experimental results showed an average detection and counting precision of 96.3% and 96.8%, respectively, which outperforms other existing approaches.

DOI： 10.3390/s19204588

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85074163412&origin=inward

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Recently, background subtraction based on deep convolutional neural networks has demonstrated excellent performance in change detection tasks. However, most of the reported approaches require pixel-level label images for training the networks. To reduce the cost of rendering pixel-level annotation data, weakly supervised learning approaches using frame-level labels have been proposed. These labels indicate if a target class is present. Frame-level supervised learning is challenging because we cannot use location information for training the networks. Therefore, some constraints are introduced for guiding foreground locations. Previous works exploit prior information on foreground sizes and shapes. In this work, we propose two constraints for weakly supervised background subtraction networks. Our constraints use binary mask images generated by simple background subtraction. Unlike previous works, our approach does not require prior information on foreground sizes and shapes. Moreover, our constraints are more suitable for change detection tasks. We also present an experiment verifying that our constraints can improve foreground detection accuracy compared to other methods, which do not include them.

DOI： 10.1109/AVSS.2019.8909896

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85076342107&origin=inward

担当区分：筆頭著者   記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

In this study, we investigated which section of a page was difficult for students to read, based on eye movement data and subjective impressions of the page's difficulty, with the aim of helping teachers revise teaching materials. It is problematic to manually model relationships between eye movements and subjective impressions of the page's difficulty. Therefore, in this study, we used a neural network to model the relationships automatically. Our method generated relevance maps representing locations where students found difficulty, in order to visualize region-wise page difficulty. To evaluate the quality of the relevance maps, we compared them with a distribution of gaze points and highlights added by the students. In addition, we administered a questionnaire to evaluate whether the relevance maps were useful to teachers when revising teaching materials. Results imply that our method can provide useful information for teachers making revisions to teaching materials.

DOI： 10.33965/celda2019_201911l014

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85078961920&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

Thanks to an increase in the amount of information on the Internet and the spread of ICT-supported educational environments, much attention has been paid to learning support based on "smart" recommendation technologies. In this study, we propose an education improvement model based on the recommender system using the human-in-the-loop design strategy. Our proposed model enhances not only learners via recommendation, but also teachers and the system itself through the interaction between teachers and the system. In this paper, we introduce the details of the proposed model and implementation strategy followed by a report of preliminary experimental results.

DOI： 10.33965/celda2019_201911l044

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85078999665&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

With the development of green energy and its means of production, more and more companies chose to build solar panel farms. However, those technologies remain relatively expensive to maintain, and prone to damages (due to natural hazards, or internal defects). Since any kind of damage on a panel cell drastically reduce a panel's efficiency, solar panels must be kept under tight supervision. With more solar panel that must be checked for damage relatively often, a cheap, accurate and fast way to find those damages must be settled. Some processes have been developed to identify panels in a true color image [1], and various ways to identify defective panels exist through image processing [2], [3] or other ways [4]. On another hand, handmade features suggest the input data obeys to some specific conditions (color, illumination), and small changes can impact accuracy. CNN [5], however, can be trained to face such changes with the appropriate dataset, and therefore be more resilient. They represent a reliable solution for identification and classification of complex features [2], [6], and can be improved more easily than handmade feature detection. In this paper is detailed the pipeline of such process, combining the straightforward approach of handmade feature detection for preprocessing to reduce the input's complexity, with the resilience of neural networks for the final identification. Detailed explanations for the different steps of the process are given: Dataset acquisition, preprocessing, and finally classification. The various leads that were followed to improve the quality of the results are also given, before comparing results with a previously used handmade detection process, and finally proposing a web user interface to exploit this process, and enrich its dataset.

DOI： 10.1117/12.2522098

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85070214379&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

This study aimed to cluster learners based on the structures of the knowledge maps they created. Learners drew their own knowledge maps to reflect their learning activities. Our system collected individual knowledge maps from many learners and clustered them to generate an integrated version of the knowledge maps of each cluster. We applied the graph analysis method to extract important keywords from the knowledge map. The results of the analysis showed that the utilization of the knowledge map helped to improve lectures and grasp the learners' level of understanding. We conducted surveys asking course managers to evaluate the effectiveness of the integrated knowledge maps of learners included in the cluster and received both positive and negative responses.

DOI： 10.33965/celda2019_201911l045

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85078968138&origin=inward

記述言語：英語   掲載種別：研究論文（国際会議プロシーディングス）

This paper introduces advanced tools in the digital learning management system M2B. The M2B system is used in Kyushu University, Japan, and contains three sub-systems: the e-learning system Moodle, the e-portfolio system Mahara, and the e-book system BookRoll. We developed useful tools to help improve both teaching and learning.

DOI： 10.1007/978-3-030-21935-2_32

Scopus

その他リンク： https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85069648388&origin=inward