Public research, which requires large computational resources, utilizes the supercomputers of the National Supercomputing Center in the Republic of Korea. The average utilization rate of resources over the past three years reached 80%. Therefore, to ensure the operational stability of this national infrastructure, specialized centers have been established to distribute the computational demand concentrated in the national centers. It is necessary to predict the computational demand accurately to build an appropriate resource scale. Therefore, it is important to estimate the inflow and outflow of computational demand between the national and specialized centers to size the resources required to construct specialized centers. We conducted a logit model analysis using the probabilistic utility theory to derive the preferences of individual users for future supercomputer resources. This analysis shows that the computational demand share of specialized centers is 59.5%, which exceeds the resource utilization plan of existing specialized centers.

Several companies have migrated their existing monolithic web applications to microservice architectures. Consequently, research on the identification of microservices from monolithic web applications has been conducted. Meanwhile, the use-case model plays a crucial role in outlining the system's functionalities at a high level of abstraction, and studies have been conducted to identify microservices by utilizing this model. However, previous studies on microservice identification utilizing use-cases did not consider the components executed in the presentation layer. Unlike existing approaches, this paper proposes a technique that considers all three layers of web applications (presentation, business logic, and data access layers). Initially, the components used in the three layers of a web application are extracted by executing all the scenarios that constitute its use-cases. Thereafter, the usage rate of each component is determined for each use-case and the component is allocated to the use-case with the highest rate. Then, each use-case is realized as a microservice. To verify the proposed approach, microservice identification is performed using open-source web applications.

Diagnosing reading disorders involves complex procedures to evaluate complex cognitive processes. For an accurate diagnosis, a series of tests and evaluations by human experts are required. In this study, we propose a quantitative tool to diagnose reading disorders based on natural reading behaviors using minimal human input. The eye movements of the third- and fourth-grade students were recorded while they read a text at their own pace. Seven machine learning models were used to evaluate the gaze patterns of the words in the presented text and classify the students as normal or having a reading disorder. The accuracy of the machine learning-based diagnosis was measured using the diagnosis by human experts as the ground truth. The highest accuracy of 0.8 was achieved by the support vector machine and random forest classifiers. This result demonstrated that machine learning-based automated diagnosis could substitute for the traditional diagnosis of reading disorders and enable large-scale screening for students at an early age.

With the rapid development of domestic and international over-the-top markets, a large amount of video content is being created. As the volume of video content increases, consumers tend to increasingly check data concerning the videos before watching them. To address this demand, video summaries in the form of plot descriptions, thumbnails, posters, and other formats are provided to consumers. This study proposes an approach that automatically generates posters to effectively convey video content while reducing the cost of video summarization. In the automatic generation of posters, face recognition and clustering are used to gather and classify character data, and keyframes from the video are extracted to learn the overall atmosphere of the video. This study used the facial data of the characters and keyframes as training data and employed technologies such as DreamBooth, a text-to-image generation model, to automatically generate video posters. This process significantly reduces the time and cost of video-poster production.

Micro ribonucleic acids (miRNAs) can regulate the protein expression levels of genes in the human body and have recently been reported to be closely related to the cause of disease. Determining the genes related to miRNAs will aid in understanding the mechanisms underlying complex miRNAs. However, the identification of miRNA-related genes through wet experiments (in vivo, traditional methods are time- and cost-consuming). To overcome these problems, recent studies have investigated the prediction of miRNA relevance using deep learning models. This study presents a method for predicting the relationships between miRNAs and genes. First, we reconstruct a negative dataset using the proposed method. We then extracted the feature using an autoencoder, after which the feature vector was concatenated with the original data. Thereafter, the concatenated data were used to train a long short-term memory model. Our model exhibited an area under the curve of 0.9609, outperforming previously reported models trained using the same dataset.

Audience reactions in response to remote virtual performances must be compressed before being transmitted to the server. The server, which aggregates these data for group insights, requires a distribution code for the transfer. Recently, distributed learning algorithms such as federated learning have gained attention as alternatives that satisfy both the information security and efficiency requirements. In distributed learning, no individual user has access to complete information, and the objective is to achieve a learning effect similar to that achieved with the entire information. It is therefore important to distribute interdependent information among users and subsequently aggregate this information following training. In this paper, we present a new extension technique for minimal code that allows a new minimal code with a different length and Hamming weight to be generated through the product of any vector and a given minimal code. Thus, the proposed technique can generate minimal codes with previously unknown parameters. We also present a scenario wherein these combined methods can be applied.

This study proposed an SCA advice system. It is an ontology-based recommender that provides advice on appropriate computer curricula based on the behavior of high school students. The three computer curricula at Chiang Mai Rajabhat University include computer science (CS), information technology (IT), and web programming and security (WEB). This study aims to design the ontology framework for an SCA advice system. The system considers three core ontologies: student, computer-curriculum, and advice. After analyzing student behaviors, the behavior types of CS, IT, and WEB were determined to be SB-2, SB-1, and SB-5, respectively. All subjects in these three curricula were analyzed and grouped into seven groups. Their curricula were synthesized in terms of basic skills, basic knowledge, and characteristics. Finally, advice results can be obtained by consolidating the curriculum nature of the CS, IT, and WEB curricula.

In this study, we developed a conversion core that produces power by utilizing the unused magnetic field in a switchboard. The conversion core makes it possible to utilize power that is normally wasted. The conversion core is composed of a core, filter, and battery. A prototype was installed in a switchboard to conduct tests on the output, battery storage, and output boosting of multiple batteries. Energy was harvested from the magnetic field generated by a busbar of the switchboard, and the power conversion ratio of the core yielded 1.08-1.01 mW per 1 A of bus current. Supplying this technology to the market after further R&D and commercialization is expected to greatly assist in the dissemination of energy harvesting, which has not yet spread widely to the general public.

The number of senior citizens with large bowel obstruction is steadily growing in Korea. Plain radiography was used to examine the severity and treatment of this phenomenon. To avoid examiner subjectivity in radiography readings, we propose an automatic segmentation method to identify fluid-filled areas indicative of large bowel obstruction. Our proposed method applies the Hough transform to locate suspicious areas successfully and applies the possibilistic fuzzy c-means unsupervised learning algorithm to form the target area in a noisy environment. In an experiment with 104 real-world large-bowel obstruction radiographs, the proposed method successfully identified all suspicious areas in 73 of 104 input images and partially identified the target area in another 21 images. Additionally, the proposed method shows a true-positive rate of over 91% and false-positive rate of less than 3% for pixel-level area formation. These performance evaluation statistics are significantly better than those of the possibilistic c-means and fuzzy c-means-based strategies; thus, this hybrid strategy of automatic segmentation of large bowel suspicious areas is successful and might be feasible for real-world use.

This paper introduces an edge AI-based scene-specific object detection system for long-term traffic management, focusing on analyzing congestion and movement via cameras. It aims to balance fast processing and accuracy in traffic flow data analysis using edge computing. We adapt the YOLOv5 model, with four heads, to a scene-specific model that utilizes the fixed camera's scene-specific properties. This model selectively detects objects based on scale by blocking nodes, ensuring only objects of certain sizes are identified. A decision module then selects the most suitable object detector for each scene, enhancing inference speed without significant accuracy loss, as demonstrated in our experiments.

In this study, we formulated a method that evaluates Taekwondo Poomsae performance using a series of choreographed training movements. Despite recent achievements in 3D human pose estimation (HPE) performance, the analysis of human actions remains challenging. In particular, Taekwondo Poomsae action analysis is challenging owing to the absence of time synchronization data and necessity to compare postures, rather than directly relying on joint locations owing to differences in human shapes. To address these challenges, we first decomposed human joint representation into joint rotation (posture) and limb length (body shape), then synchronized a comparison between test and reference pose sequences using DTW (dynamic time warping), and finally compared pose angles for each joint. Experimental results demonstrate that our method successfully synchronizes test action sequences with the reference sequence and reflects a considerable gap in performance between practitioners and professionals. Thus, our method can detect incorrect poses and help practitioners improve accuracy, balance, and speed of movement.

Trajectory prediction is an essential element for driving autonomous vehicles, and various trajectory prediction models have emerged with the development of deep learning technology. Convolutional neural network (CNN) is the most commonly used neural network architecture for extracting the features of visual images, and the latest models exhibit high performances. This study was conducted to identify an efficient CNN backbone model among the components of deep learning models for trajectory prediction. We changed the existing CNN backbone network of multiple-trajectory prediction models used as feature extractors to various state-of-the-art CNN models. The experiment was conducted using nuScenes, which is a dataset used for the development of autonomous vehicles. The results of each model were compared using frequently used evaluation metrics for trajectory prediction. Analyzing the impact of the backbone can improve the performance of the trajectory prediction task. Investigating the influence of the backbone on multiple deep learning models can be a future challenge.

This study designed a model-based Vehicle Control Unit (VCU) software for electric vehicles. Electric vehicles have transitioned from conventional powertrains (e.g., engines and transmissions) to electric powertrains. The primary role of the VCU is to determine the optimal torque for driving control. This decision is based on the driver's power request and current road conditions. The determined torque is then transmitted to the electric drive system, which includes motors and controllers. The VCU employs an Artificial Neural Network (ANN) and calibrated reference torque to enhance the electric vehicle's performance. The designed VCU software further refines the final reference torque by comparing the control logic with the torque calculation functions and ANN-generated reference torque. Vehicle tests confirmed the effective optimization of vehicle performance using the model-based VCU software, which includes an ANN.