As digital materials proliferate in higher education, e-book interaction logs provide a scalable lens on how students study. However, most existing research analyzes these logs in a one-dimensional manner, which limits the ability to capture students’ study approaches comprehensively. Moreover, the relationship between students’ study approaches, metacognition, and performance remains unclear. To address these challenges, we propose a three-dimensional framework that incorporates engagement, navigation pattern, and context, combining theory-driven and data-driven perspectives to define behavior-based features and group students with similar patterns. We apply this approach to data from a real-world class in which students used an e-book system to study course materials and complete comprehension quizzes. Our analysis identified three distinct groups of students with different study approaches, and revealed that engagement investment alone does not guarantee achievement. We further examined how these approaches relate to students’ metacognitive awareness and academic performance.

Traditional knowledge graph (KG) approach often rely on static textbook content and overlook the dynamic, collaborative interactions in collaborative problem-solving (CPS). This study introduced a three-step integrated KG approach designed to support CPS in STEM education and examined the effective KG features that influence CPS learning outcomes. KGs were generated by combining learning materials and student dialogue data. Two types of features, graph structural features and knowledge activation features, were identified to classify knowledge nodes and analyze how students activated knowledge during CPS. Clustering analysis revealed three types of knowledge nodes: Peripheral Nodes, Core Nodes, and Degree Hubs. Furthermore, key features such as depth, branch, and activated paths showed positive correlations with group discussion performance and CPS skills but had limited influence on test scores. These findings highlight the potential of integrated KGs to support both individual and group learning in STEM education.

Augmented reality (AR) is shifting language learning to more authentic and interactive. However, the complex and multitasking nature of AR creates challenges for learners in monitoring their progress. While Learning Analytics Dashboards (LADs) support assessment, they rarely address learning strategies in AR environments. This study first identified effective AR language learning strategies using combined Lag Sequential Analysis and Ordered Network Analysis, then developed a LAD with integrated learning prompts based on these findings. Formative evaluation confirmed the prompts of LAD effectively supported learners in monitoring and reflecting on their learning processes. This research contributes an analytical approach for identifying effective strategies and incorporating them into LAD design.

E-books generate extensive log data that sheds light on student behaviors. Among these, page jumps offer unique insights into reading strategies. However, prior research seldom connects these jumps to both the e-book's content and page functions and often presents only fragmented information. This study investigates how e-book content and page types relate to student page jump behaviors. We also propose a visualization framework that integrates e-book content and log data, enabling intuitive reading path visualizations and detailed analysis of student interactions. Our approach aims to offer educators actionable insights for refining instructional materials and providing more personalized feedback, ultimately enhancing the e-book learning experience.

Picture-book storytelling fosters preschool children's cognitive and literacy development in Japanese childcare centers. However, training childcare workers remains challenging due to the need for emotional engagement and adaptive strategies. This study develops a VR-based storytelling training system that enables trainees to practice in an immersive environment with virtual child avatars. The system evaluates performance based on voice volume, reading speed, posture, gaze, and eye contact, providing real-time feedback. An expert evaluation confirmed the system's validity but identified areas for improvement. Adaptive scoring for voice and reading speed, tracking of page-turning techniques, and body orientation assessment should be incorporated. Additionally, realistic child reactions and adjusted gaze evaluation for children sitting on the floor are needed.

Recently, Ainsworth et al. (2023) showed that using weight matching (WM) to minimize the $L^2$ distance in a permutation search of model parameters effectively identifies permutations that satisfy linear mode connectivity (LMC), where the loss along a linear path between two independently trained models with different seeds remains nearly constant. This paper analyzes LMC using WM, which is useful for understanding stochastic gradient descent's effectiveness and its application in areas like model merging. We first empirically show that permutations found by WM do not significantly reduce the $L^2$ distance between two models, and the occurrence of LMC is not merely due to distance reduction by WM itself. We then demonstrate that permutations can change the directions of the singular vectors, but not the singular values, of the weight matrices in each layer. This finding shows that permutations found by WM primarily align the directions of singular vectors associated with large singular values across models. This alignment brings the singular vectors with large singular values, which determine the model's functionality, closer between the original and merged models, allowing the merged model to retain functionality similar to the original models, thereby satisfying LMC. This paper also analyzes activation matching (AM) in terms of singular vectors and finds that the principle of AM is likely the same as that of WM. Finally, we analyze the difference between WM and the straight-through estimator (STE), a dataset-dependent permutation search method, and show that WM can be more advantageous than STE in achieving LMC among three or more models.

Ainsworth et al. empirically demonstrated that linear mode connectivity (LMC) can be achieved between two independently trained neural networks (NNs) by applying an appropriate parameter permutation. LMC is satisfied if a linear path with non-increasing test loss exists between the models, suggesting that NNs trained with stochastic gradient descent (SGD) converge to a single approximately convex low-loss basin under permutation symmetries. However, Ainsworth et al. verified LMC for two models and provided only limited discussion on its extension to multiple models. In this paper, we conduct a more detailed empirical analysis. First, we show that existing permutation search methods designed for two models can fail to transfer multiple models into the same convex low-loss basin. Next, we propose a permutation search method using a straight-through estimator for multiple models (STE-MM). We then experimentally demonstrate that even when multiple models are given, the test loss of the merged model remains nearly the same as the losses of the original models when using STE-MM, and the loss barriers between all permuted model pairs are also small. Additionally, from the perspective of the trace of the Hessian matrix, we show that the loss sharpness around the merged model decreases as the number of models increases with STE-MM, indicating that LMC for multiple models is more likely to hold. The source code implementing our method is available at https://github.com/e5-a/STE-MM.

This study investigates the potential of using virtual reality (VR) technology to induce emotions that correspond to Japanese psych-mimetic words. Considering the complexities of learning these words, previous work [1] has developed a VR system tailored for psych-mimetic word education. By crafting immersive scenes that evoke specific emotions, the system is designed to help learners internalize the feelings associated with these words. Therefore, it is vital to confirm the system’s effectiveness in eliciting emotional responses from learners, through formative evaluation combining eye-tracking data analysis from VR headsets with semi-structured interviews conducted in the VR scenes. From both objective and subjective perspectives, this study investigates whether participants can experience emotions that align with the psych-mimetic words portrayed in the VR scenes. The results suggest that participants can generate the intended emotions within these VR scenes. This study intends to conduct future studies further examine the system’s impact on the educational outcomes of Japanese psychmimetic words. The current findings offer robust empirical evidence supporting the use of VR technology in language education.

As information technology advanced, accuracy of technology-driven assessment is being improved. In order to assess learning performance and awareness, assessment of metacognition level with technology can be useful to understand learner’s learning comprehension and awareness. Metacognition is one of the most important elements for successful learning. However, current way to evaluate metacognition level focuses on psychological method such as questionnaire and interview. The recent growth of learning analytics research has demonstrated the relationships between metacognition, learning awareness, and learning behaviors. This study aims to investigate metacognitive learning behaviors using small grain data on eBook and learning analytics dashboard (LAD) over eight weeks in a university course. To do so, we determined high and low metacognitive learner groups using the Metacognitive Awareness Inventory and investigated the differences between the two groups in eBook and LAD. The findings suggest that four learning behaviors eBook and LAD were detected as metacognitive learning behaviors, and contribute to the improvement of technology-driven assessment.

Adversarial training (AT) is a promising method to improve the robustness against adversarial attacks. However, its performance is not still satisfactory in practice compared with standard training. To reveal the cause of the difficulty of AT, we analyze the smoothness of the loss function in AT, which determines the training performance. We reveal that nonsmoothness is caused by the constraint of adversarial attacks and depends on the type of constraint. Specifically, the$L_\infty$constraint can cause nonsmoothness more than the$L_2$constraint. In addition, we found an interesting property for AT: the flatter loss surface in theinput spacetends to have the less smooth adversarial loss surface in theparameter space. To confirm that the nonsmoothness causes the poor performance of AT, we theoretically and experimentally show that smooth adversarial loss by EntropySGD (EnSGD) improves the performance of AT.

This paper proposes a new loss function for adversarial training. Since adversarial training has difficulties, e.g., necessity of high model capacity, focusing on important data points by weighting cross-entropy loss has attracted much attention. However, they are vulnerable to sophisticated attacks, e.g., Auto-Attack. This paper experimentally reveals that the cause of their vulnerability is their small margins between logits for the true label and the other labels. Since neural networks classify the data points based on the logits, logit margins should be large enough to avoid flipping the largest logit by the attacks. Importance-aware methods do not increase logit margins of important samples but decrease those of less-important samples compared with cross-entropy loss. To increase logit margins of important samples, we propose switching one-vs-the-rest loss (SOVR), which switches from cross-entropy to one-vs-the-rest loss for important samples that have small logit margins. We prove that one-vs-the-rest loss increases logit margins two times larger than the weighted cross-entropy loss for a simple problem. We experimentally confirm that SOVR increases logit margins of important samples unlike existing methods and achieves better robustness against Auto-Attack than importance-aware methods.

The variational autoencoder (VAE) is a powerful latent variable model for unsupervised representation learning. However, it does not work well in case of insufficient data points. To improve the performance in such situations, the conditional VAE (CVAE) is widely used, which aims to share task-invariant knowledge with multiple tasks through the task-invariant latent variable. In the CVAE, the posterior of the latent variable given the data point and task is regularized by the task-invariant prior, which is modeled by the standard Gaussian distribution. Although this regularization encourages independence between the latent variable and task, the latent variable remains dependent on the task. To reduce this task-dependency, the previous work introduced an additional regularizer. However, its learned representation does not work well on the target tasks. In this study, we theoretically investigate why the CVAE cannot sufficiently reduce the task-dependency and show that the simple standard Gaussian prior is one of the causes. Based on this, we propose a theoretical optimal prior for reducing the task-dependency. In addition, we theoretically show that unlike the previous work, our learned representation works well on the target tasks. Experiments on various datasets show that our approach obtains better task-invariant representations, which improves the performances of various downstream applications such as density estimation and classification.

We propose a method for improving adversarial robustness by addition of a new bounded function just before softmax. Several studies hypothesize that small logits (inputs of softmax) by logit regularization contributes to adversarial robustness of deep learning. Following this hypothesis, we analyze norms of logit vectors at the optimal point under the assumption of universal approximation and explore new methods for constraining logits by addition of a bounded function before softmax. We theoretically and empirically reveal that small logits by addition of a common activation function, e.g., hyperbolic tangent, do not improve robustness since input vectors of the function (pre-logit vectors) can have large norms. From the theoretical findings, we develop the new bounded function. The addition of our function contributes to adversarial robustness because it makes logit and pre-logit vectors have small norms. Since our method only adds one activation function before softmax, it is easy to combine our method with adversarial training. Our experiments demonstrate that our method is comparable to logit regularization methods in terms of robustness against untargeted attacks without adversarial training. Furthermore, it is superior or comparable to logit regularization methods and a recent defense method (TRADES) when using adversarial training.

We propose the factorizing variational autoencoder (FAVAE), a generative model for learning dis- entangled representations from sequential data via the information bottleneck principle without supervision. Real-world data are often generated by a few explanatory factors of variation, and disentangled representation learning obtains these factors from the data. We focus on the disen- tangled representation of sequential data which can be useful in a wide range of applications, such as video, speech, and stock markets. Factors in sequential data are categorized into dynamic and static ones: dynamic factors are time dependent, and static factors are time independent. Previous models disentangle between static and dynamic factors and between dynamic factors with different time dependencies by explicitly modeling the priors of latent variables. However, these models cannot disentangle representations between dynamic factors with the same time dependency, such as disentangling “picking up” and “throwing” in robotic tasks. On the other hand, FAVAE can disentangle multiple dynamic factors via the information bottleneck principle where it does not require modeling priors. We conducted experiments to show that FAVAE can extract disentangled dynamic factors on synthetic, video, and speech datasets.

Summer vacation is considered a cause of loss in students' learning performance. In this study, we investigated the differences in learning behaviors in reading learning materials and time-related behavior patterns regarding summer assignments among three classes under different instructional styles. The results showed that students' learning behaviors in summer were correlated with instructional styles in school.

Previous research has established that motivation has a positive impact on the learning processes and behaviors [1] [2]. Learning analytics (LA) can play an important role in addressing the issue of collecting learning behaviors. In this study, we observed the teaching activities of three classes and examined the relationships among learning motivation, learning performance, and learning behaviors of students in digital learning material readers.

One problem in real-world applications of reinforcement learning is the high dimensionality of the action search spaces, which comes from the combination of actions over time. To reduce the dimensionality of action sequence search spaces, macro actions have been studied, which are sequences of primitive actions to solve tasks. However, previous studies relied on humans to define macro actions or assumed macro actions to be repetitions of the same primitive actions. We propose encoded action sequence reinforcement learning (EASRL), a reinforcement learning method that learns flexible sequences of actions in a latent space for a high-dimensional action sequence search space. With EASRL, encoder and decoder networks are trained with demonstration data by using variational autoencoders for mapping macro actions into the latent space. Then, we learn a policy network in the latent space, which is a distribution over encoded macro actions given a state. By learning in the latent space, we can reduce the dimensionality of the action sequence search space and handle various patterns of action sequences. We experimentally demonstrate that the proposed method outperforms other reinforcement learning methods on tasks that require an extensive amount of search.

The variational autoencoder (VAE) is a powerful generative model that can estimate the probability of a data point by using latent variables. In the VAE, the posterior of the latent variable given the data point is regularized by the prior of the latent variable using Kullback Leibler (KL) divergence. Although the standard Gaussian distribution is usually used for the prior, this simple prior incurs over-regularization. As a sophisticated prior, the aggregated posterior has been introduced, which is the expectation of the posterior over the data distribution. This prior is optimal for the VAE in terms of maximizing the training objective function. However, KL divergence with the aggregated posterior cannot be calculated in a closed form, which prevents us from using this optimal prior. With the proposed method, we introduce the density ratio trick to estimate this KL divergence without modeling the aggregated posterior explicitly. Since the density ratio trick does not work well in high dimensions, we rewrite this KL divergence that contains the high-dimensional density ratio into the sum of the analytically calculable term and the lowdimensional density ratio term, to which the density ratio trick is applied. Experiments on various datasets show that the VAE with this implicit optimal prior achieves high density estimation performance.

Game-based learning for informal learning has become an issue in digital game-based learning research. However, assessments from the observations of the learning process are difficult in the non-face-to-face situation of an informal settings. This research aims to evaluate the effects of a game-based informal learning system for history, “Hist Maker ” integrating the external assessment with tests and the game-embedded assessment with the analysis of players’ gameplay log data. For the data analysis, we integrated the statistical model and learning analytics technology through cluster analysis. This approach allowed us to draw conclusions about the correlation between players’ behavior patterns and learning effects in the game. These conclusions show the potential of this approach to solve the observation problem in research on serious games for informal learning.

Extended learning support systems for all-age education requires inclusive user interface design, especially for elderly users. A dual-tablet user interface with simplified visual layers and more intuitive operations was proposed aiming to reduce the physical and mental loads of elderly learners. An initial prototype with basic functions of viewing learning material was developed based on a cross-platform framework. Two preliminary user experiments participated by elderly volunteers were carried out for formative evaluations, in order to improve the usability of the interface design iteratively. The prototype was modified based on the participants' comments and observation of their operations during the experiments. Additional findings of the elderly users' preference and tendency were discussed for further development.

To improve designs of e-learning materials, it is necessary to know which word or figure a learner felt "difficult" in the materials. In this pilot study, we measured electroencephalography (EEG) and eye gaze data of learners and analyzed to estimate which area they had difficulty to learn. The developed system realized simultaneous measurements of physiological data and subjective evaluations during learning. Using this system, we observed specific EEG activity in difficult pages. Integrating of eye gaze and EEG measurements raised a possibility to determine where a learner felt "difficult" in a page of learning materials. From these results, we could suggest that the multimodal measurements of EEG and eye gaze would lead to effective improvement of learning materials. For future study, more data collection using various materials and learners with different backgrounds is necessary. This study could lead to establishing a method to improve e-learning materials based on learners' mental states.

Online learning environments presently accumulate large amounts of log data. Analysis of learning behaviors from these log data is expected to benefit instructors and learners. This study was intended to identify effective measures from e-book materials used at Kyushu University and to employ these measures for analyzing learning behavioral patterns. In an evaluation, students were grouped into four clusters using k-means clustering, and their learning behavioral patterns were analyzed. We examined whether the learning behavioral patterns exhibited relations with the learning outcomes. The results reveal that the learning behavior of “backtrack” style reading exerts a significant positive influence on learning effectiveness, which can aid students to learn more efficiently.

We propose a robust multivariate density estimator based on the variational autoencoder (VAE). The VAE is a powerful deep generative model, and used for multivariate density estimation. With the original VAE, the distribution of observed continuous variables is assumed to be a Gaussian, where its mean and variance are modeled by deep neural networks taking latent variables as their inputs. This distribution is called the decoder. However, the training of VAE often becomes unstable. One reason is that the decoder of VAE is sensitive to the error between the data point and its estimated mean when its estimated variance is almost zero. We solve this instability problem by making the decoder robust to the error using a Bayesian approach to the variance estimation: we set a prior for the variance of the Gaussian decoder, and marginalize it out analytically, which leads to proposing the Student-t VAE. Numerical experiments with various datasets show that training of the Student-t VAE is robust, and the Student-t VAE achieves high density estimation performance.

This preliminary research investigates the effects of self-regulated learning support using the system "Self-regulator (SR)," and relationships between self-regulated learning awareness, learning behaviors, and perceived effects of SR. The results showed that the course with SR promoted "meet the deadline" awareness. The results of Spearman's correlation analysis revealed that procrastination awareness for high performance is one of the key factors for time management, which is an important factor of self-regulated learning.

In this paper, we examined whether previous findings on educational big data consisting of e-book logs from a given academic course can be reproduced with different data from other academic courses. The previous findings showed that (1) students who attained consistently good achievement more frequently browsed different e-books and their pages than low achievers and that (2) this difference was found only for logs of preparation for course sessions (preview), not for reviewing material (review). Preliminarily, we analyzed e-book logs from four courses. The results were reproduced in only one course and only partially, that is, (1) high achievers more frequently changed e-books than low achievers (2) for preview. This finding suggests that to allow effective usage of learning and teaching analyses, we need to carefully construct an educational environment to ensure reproducibility.

This research aims to investigate the relationship between self-regulated learning awareness, learning behaviors, and learning performance in ubiquitous learning environments. In order to do so, psychometric data about self-regulated learning and log data such as marker, annotation, accessing device types that stored the learning management system were collected and analyzed using multiple regression analysis with stepwise method. The results indicated that self-efficacy, internal value, and the number of read slides had a significant influence on the final score, and the awareness of cognitive learning strategy use has slightly significant power to predict the final score.

This paper discusses learning behavior analysis using a learning management system (LMS) and an e-textbook system. We collected a large number of operation logs from e-textbooks to analyze the process of learning. In addition, we conducted a quiz to check the level of understanding. In our study, we especially focus on an analysis of the relationship between learning behavior in informal learning and its effectiveness in the corresponding quiz. We apply a machine learning and classification methodology for behavior analysis. Our experimental results demonstrate that students who undertake good informal learning achieve better scores in quizzes.

This preliminary research investigates the relationship between psychometric data and learning behaviors in the learning analytics research field, specifically, the relationship between self-regulated learning and learning behavior. The results of this limited research show that marker and annotation use have a weak significant relationship with self-efficacy and the intrinsic value of learning materials.

The purpose of this research is to report university students’ learning experiences with information communication technology (ICT) and open educational sources (OES) for global learning in Japan. The survey of 327 Japanese university students included seven multiple-choice items and 16 open-ended questions about students’ learning experiences. The results showed that the most frequent use of ICT, including computer-mediated communication, is a discussion function of Blackboard for formal collaborative learning and LINE for informal learning. Learning with OES is less popular; only one student had taken a two-month course via YouTube provided by a Japanese university, and only two students had learned with foreign students online. The students’ preferred activities for future international collaborative learning included project-based learning, casual chats and conversation, discussion, and e-mail exchange. Favored topics were ones related to their majors, international situations, the environment, cultural differences, and school life. Language proficiency, communication, cultural differences, and values and beliefs caused the most anxiety and concern for international collaboration, but the time gap, legal issues, and infrastructure were also considered.

Nowadays, it is a hot topic to analyze the huge amount of data in the world. This issue also exists in the learning during students’ life. The learning data are collected only to record students’ learning status. As a result, most learning data are not used to improve the quality of learning for students. In this paper, we propose an order made education system, which can recommend students to select the courses they want to learn. In order to analyze students’ learning behaviors, we collect students’ learning data by using mobile devices.

The purposes of this preliminary study are (1) to investigate possible factors for English as a foreign language (EFL) learners’ behavior and attitudes towards computer-supported collaborative learning focusing on online discussion. Comment numbers, satisfaction and perceived group contributions on two online discussions of 58 EFL students were analyzed with multiple regression correlation, in relation with items of six inventories; (a) social skill (Kikuchi, 2007), (b) Self-efficacy for English learning with CSCL based on Matsunuma (2006), (c) English learning strategy (Kubo, 1999), (d) Social presence and cognitive presence with text-chat developed based on Gunawardena & Zittle (1997), (e) Felder-Soloman Index of Learning Style (ILS) (Felder & Silverman, 1988), and (f) researcher-developed questionnaire for a discussion task. The results show that 11 items collectively account for 76.2% of the comment number variance (F(11,46)=13.39, p<.01), 15 items and “sequencel” type score of learning style significantly explain 89.2% of the learners’ satisfaction (F(16, 41)=21.12, p<.01), and 2 items and “sequential” type score of learning style describe 32.8% of the learners’ perceived contribution (F(3, 54)=8.80, p<.01). There are three overlapped items and one learning style type for all prediction equations. It indicates that 25 items and 11 items for sequential/global type in the learning style types may be utilized to predict EFL learners’ behavior and attitudes towards online discussion.

This paper discusses the configuration of an Internet-based system in which products and services relating to digital content are offered to users by employing the ISO MPEG-21 standard and the de facto standards: Content ID (defined by cIDf) and DOI (defined by IDF).