Earticle

Breast cancer remains the foremost cause of cancer-related mortality worldwide. The histopathological diagnosis is impeded by the intricate nature of image interpretation and the presence of inter-observer variability among pathologists. Deep learning (DL) for cancer image understanding has revolutionized accurate breast cancer diagnosis, marking a significant advancement in medical image analysis. Researchers proposed DL-based intelligent models to overcome the challenges of manual observations. However, the existing models suffer from a considerable computational burden, demanding substantial time investments that restrict efficient and scalable breast cancer diagnosis solutions. Our study introduces an automated breast cancer diagnosis system employing a lightweight Convolutional Neural Network (CNN) model, adept at extracting intricate features from histopathological images. Our system has attained superior accuracy through extensive experimentation on a comprehensive breast cancer dataset while employing fewer parameters compared to state-of-theart (SOTA) techniques.

In the era of digital transformation, various attempts are constantly being made to foster software talent required to lead the future society. Accordingly, educational research is being conducted to develop computational thinking, a thinking process that uses the basic concepts of computers to solve problems. In particular, various assessment tools are being used to diagnose computational thinking. Typically, online judge systems that can solve programming problems and automatically evaluate programming code in an online environment are used, and self-report type assessment questions are used to diagnose computational thinking. However, existing online judge systems do not provide questions to diagnose computational thinking, and only present the correctness of the answer based on the execution result of the code. In addition, the self-assessment type of computational thinking assessment tool has the limitation that it is unreliable because they are subjective and lack objectivity. In this study, we develop a question instrument that can diagnose computational thinking through a different approach than the existing computational thinking assessment tools. We develop a problem context-driven item and then verify it through a data analytics-driven scheme using an online judge system. In particular, to diagnose computational thinking, we present problem situations and provide items for each specific element of computing thinking by categorizing each situation into abstraction, algorithm, and automation. In addition, based on the programming code written in the automation step, the answers in the abstraction and algorithm section are checked so that data analytics-driven question modification and verification can be made. This study provides a new direction for diagnosing computational thinking and aims to provide learners with more reliable diagnostic results.

Customer Churn Prediction is the process of identifying customers who are likely to stop using a company's products or services in the near future that is critical for the long-term financial stability of a business. Retaining existing customers is often more cost-effective than acquiring new ones, making churn prediction a key focus for customer relationship management (CRM). This study aimed to identify customer churn in data source containing 8,047 sale transactions with various features such as sales, profit, and product category. The four techniques of churn labels generation were introduced base-on features of Time, Value, and Feedback. Additionally, a combination of Time and Value also used to test. The data was split into 80% training and 20% testing subsets, focusing on seven selected features and the Multiple Criteria churn label. Four machine learning models—Random Forest (RF), Logistic Regression (LR), Gradient Boosting (GB), and Support Vector Machines (SVM) were used to create model. The results showed that LR (73.60%) and SVM (71.70%) performed a good performance in terms of accuracy. However, to compare with dataset that included churn label as E-commerce Customer Behavior and Purchase Dataset [23] the results showed that the proposed techniques can be used to impute a churn label attribute which effecting to a classification model.

This paper delves into the crucial need for robust security measures to protect web services from cyber threats that can compromise their availability, veracity, and privacy. The introduction highlights the inadequacy of existing solutions that are limited to specific use areas and emphasizes the importance of web services being able to respond effectively to assaults and incursions. The related work section discusses the research on web services that has primarily focused on modeling, design, and testing in light of more damaging threats. The methodology section outlines the penetration testing cycle for the SPARCI System, which includes testing for various types of attacks such as DOS, Brut Force Web, Brute Force SSH, Session Hijacking, DNS Spoofing, ARP poisoning, NoSQL Injection, and Memory dump. In this study, we have worked on DOS attacks, and the remaining will be targeted later in the future. The experimental setup includes the necessary instance, Linux system, Python packages, and installed software such as Burp Suite to perform the attack. The results section presents the findings of the penetration testing and the effectiveness of the mitigation strategies employed. The paper concludes that the proposed methodology can be used to identify vulnerabilities in web services and provide recommendations for their mitigation. The study highlights the importance of continuous monitoring and testing of web services to ensure their security and prevent cyber-attacks in terms of identifying and mitigating various vulnerabilities in the existing systems.as.

In autonomous driving, environmental perception and decision-making are technologies that acquire signals based on various sensors and generate information that enables obstacle avoidance, emergency stops, and path planning. Such environmental perception technologies are limited in that they depend on expensive sensors such as LIDAR and RADAR, so research on environmental perception technologies using only cameras is actively being conducted. Studies predicting traffic accidents based on dashcam footage have also been performed as part of such research. This is challenging because only limited forward-looking footage can be acquired, and the surrounding environment is dynamic and changes quickly, making analysis difficult. Existing research has focused on learning the spatialtemporal feature representation to solve these problems. This paper extracts the trajectories of the ego-vehicle and surrounding vehicles using Visual SLAM and Multiple Object Tracking algorithms and uses them as inputs to the graph convolutional neural network(GCN) to learn the spatialtemporal feature representation. In addition, the features learned through the GCN are used as inputs to a bayesian neural network(BNN) to predict the probability of accidents, and its ability to predict accidents in advance has been verified by comparison with existing studies.

EEG signals have been widely used in emotion recognition in recent years. However, a great challenge still exists for the practical applications of cross-subject emotion recognition. Inspired by recent neuroscience studies and the advantage of the DE feature applied in EEG emotion recognition, we proposed a combined DE feature and contrastive learning method to tackle the cross-subject emotion recognition problem. The proposed model can minimize the inter-subject differences by maximizing the similarity in EEG signal representations across subjects when they receive the same emotional stimuli in contrast to different ones and gain a better encoding. Finally, we conducted extensive experiments on SEED and SEED-IV. The cross-subject emotion recognition accuracy is 84.72 on the SEED and 69.24 on the SEED-IV. It experimentally verified the effectiveness of the model.

The 2D and 3D pose estimation methods have now improved well in general performance but have not yet been emphasized in terms of speed and efficiency for the infant dataset and the existence of public data on infants is a significant challenge. Furthermore, clinical studies related to the analysis of the pose and movements of infants are attracting considerable attention. That motivated us to collect infant data and develop a lighter model for estimating infant poses that can run on edge devices and CPUs. Most current methods are characterized by complex structures and multiple parallel branches of inference to synthesize pose estimated results. In this project, we aim to refine the architecture of the pose estimation algorithm based on an approach of OpenPose-2016, for use on edge devices and training that model on 2D images. The proposed simplified model features a single-branch structure designed to estimate infant pose with a size of 4.09 million parameters. The model when executed undergo algorithmic complexity of 8.97 giga floating point operations per second (GFLOPS), allowing it to run at approximately 23 frames per second on a Core i5-10400f. The proposed methodology demonstrates compact dimensions while achieving superior performance compared to existing methods on the same self-collected infant dataset. It is hoped that this straightforward and pragmatic approach will establish a robust foundation and provide favorable conditions for future research in the application of pose estimation.

In recent object detection research, there has been a growing focus on Detection Transformers predicting bounding boxes directly. However, Detection Transformers face challenges such as slow convergence and difficulty in detecting small objects. We attribute these issues to the insufficient feature extraction capability of the backbone. Therefore, we employ the high-performing backbone, the Pyramid Pooling Transformer to detection Transformer. However, we observe a problem where, despite rapid initial convergence, the model fails to converge effectively after a certain point in training. We discuss the underlying causes of this issue in this study.

Although indoor positioning technologies have been extensively researched, many indoor positioning methods are limited by the need for actual measurement data to be collected within a specified environment. To address issues, this paper proposes an indoor positioning simulation to implement an indoor environment and measure object position. In Addition, it presents the architecture and method of simulation through simple situations.

This article addresses an energy-efficient positioning system using Ultra-Wideband (UWB) technology. Positioning systems play a crucial role in various applications such as disaster rescue, autonomous robots, and industrial automation. However, in such environments, IoT devices rely on batteries, and UWB systems have been limited in energy efficiency compared to other wireless communication technologies. In this study, we propose a high-resolution positioning algorithm for UWB systems with reduced communication frequency to enhance energy efficiency while providing real-time location information. Experimental results demonstrate that reducing the UWB communication frequency leads to increased energy efficiency. We also analyze the relationship between real-time location update speed and the energy constraints of IoT devices. Finally, we compare the performance of UWB systems with the high-resolution algorithm to conventional triangulation methods. This research provides a fundamental basis for improving energy-efficient positioning systems using UWB technology.

Artificial Intelligence (AI) has demonstrated unprecedented performance across a multitude of sectors, including disaster response. However, deploying AI in safetycritical environments poses unique challenges, especially regarding thermal management and real-time decision-making. Utilizing Edge TPU, one of the Neural Processing Units (NPUs), has shown promise in overcoming some of these challenges. Despite its advantages, Edge TPU still has limitations in thermal management and real-time task scheduling. This study introduces an approach employing Dynamic Frequency Scaling (DFS) and SRAM allocation techniques to address these challenges. By dynamically adjusting operating frequencies and resource allocations, the proposed approach aims to optimize both thermal management and real-time performance, thereby enhancing the reliability and efficiency of AI technologies in critical applications like disaster response.

Over the past few decades, home automation systems have gained significant popularity due to their ability to enhance comfort and improve the quality of life. However, with the increasing reliance on internet connectivity, these systems face challenges in coping with the expanding attack surfaces and the corresponding attacks. This paper provides an overview of the attacks that commonly target smart home devices and networks. It explores the vulnerabilities that can be exploited by attackers to compromise the security and privacy of these systems. Furthermore, the paper proposes and discusses potential solutions and countermeasures to mitigate these attacks. By understanding the nature of these threats and implementing appropriate security measures, homeowners and system developers can enhance the security posture of smart home systems, ensuring a safer and more secure environment.

The Metaverse, a rapidly evolving digital realm where people connect, collaborate, and engage in various activities, presents promising opportunities and, at the same time, raises significant security concerns. As users interact with diverse services, the need for a dependable method to confirm user identities becomes apparent. In light of current technology and the capabilities of biometrics, this paper explores the application of biometric authentication as a robust security solution within the Metaverse. Specifically, it investigates the practicality and effectiveness of biometric authentication methods, including facial recognition and iris scan, in preserving the integrity, confidentiality, and authenticity of interactions in the Metaverse. To illustrate our findings, we propose an authentication mechanism tailored for the Metaverse in response to the security challenges. To validate the feasibility and benefits of our approach, we conducted evaluations within an Apple Mac Mini environment, implementing biometric authentication in a simulated Apple Vision Pro. The outcomes of our study conclusively illustrate the security enhancements and advantages of our proposed authentication mechanism.

Satellite navigation systems are used in various fields requiring positioning and navigation. They transmit navigation messages including timing and orbit information through signals from satellites. However, the signals can be threatened by spoofing attacks that forge the satellite signals. To address this threat, satellite systems provide authentication services for navigation messages. In this paper, we analyze the signal authentication protocol of the Japanese satellite navigation system, QZSS. We also compare its authentication performance with those of GPS, Galileo, and BeiDou.

The in-vehicle infotainment (IVI) system provides driving-related information such as vehicle status and route guidance, as well as entertainment services for users. In other words, the IVI system performs a variety of functions such as audio and video data streaming, navigation, smartphone pairing and mirroring, and voice-activated vehicle control. Therefore, the IVI system stores a very large amount of log data, including fuel consumption, current vehicle status, navigation information, as well as information about external devices (e.g., smartphones, tablets) connected to it. Manually analyzing such log data is cumbersome and time-consuming. To solve this problem, this paper proposes a method to efficiently analyze log data of the IVI system. The proposed method can automatically extract and analyze important information, enabling effective vehicle digital forensic investigation.

Crowdsourcing has been widely applied to collect or label data through the Internet. In crowdsourcing, truth discovery enables servers to estimate and extract truthful information from multiple possibly conflicting data sources. However, apply crowdsourcing to fog computing still suffers from one critical problem: How to achieve privacy preservation under a flexible fog-computing architecture without the assistance of a cloud. Collusion and fault tolerance of fog server's dropout should be carefully taken care. A privacypreserving crowdsourcing scheme with truth discovery in fog computing is proposed. The scheme preserves privacy of workers’ data and satisfies the (T, N)-threshold property, which resists collusion of T-1 number of fog servers and 𝑵 − 𝑻 number of dropouts of the servers. Extensive analysis is conducted on privacy preservation and reliability, showing that the proposed scheme is effective and secure in fog computing. Experimental results show that the proposed scheme is efficient regarding computation and network communication.

This study presents an innovative approach that utilizes the immersive capabilities of the Oculus Quest to create a rehabilitation environment in the form of a game for individuals with paraplegia. By combining aspects of virtual reality, gamification, and reinforcement learning, this research establishes a customized framework tailored to meet the unique needs of paraplegic patients. The research encompasses various stages including the development of a virtual reality gamified setting, the implementation of a reinforcement learning agent, data collection methods, and aims to revolutionize paraplegic rehabilitation by offering an interactive and individualized experience.

An important factor in virtual reality (VR) is not only technological advancement but also understanding VR users to provide a high user experience. The collection and analysis of VR sensor signals generated by users play a role in understanding user states and behaviors. In this paper, we present three VR case studies that involve the collection and analysis of sensor signals generated during the VR experience. We discuss the importance of using multimodal VR sensor data by analyzing sensor signals generated by users in the case studies.

Predicting disaster classification on time is critical to mitigate the damage. Since identifying disaster types requires large amounts of data, and real-world data are often imbalanced, there are many recent works addressing data imbalance problems using generative models. However, if the process of generating text data based on disaster classes and severity is not handled improperly, the quality of the data can be degraded as well as the performance of classification predictions. In this paper, we propose a scheme for generating data with enhanced quality using text based on labels such as informational value of text and severity of disasters. Our experiment results verify the quality of data through the comparisons of prediction performance between various machine learning models.

This work presents a novel fine-tuning scheme for enhancing the quality of Subject Driven Image Generation. Motivated by recent works on fine-tuning pre-trained diffusion models, we extract information from visual patch embedding to optimize the performance of the image encoder in our proposed method. Additionally, the loss function of the conventional Unet model is replaced with Masked Diffusion Loss. During inference time, the model can control degree of similarity between result image and reference image by using Classifier - Free Guidance method. Experimental results indicate that the proposed model exhibits improved image generation quality in comparison to the previous schemes.

In the modern society, with the development of Information & Communications Technology (ICT), the cyber threat is also increasing, and to prepare for this, Moving Target Defense (MTD) strategy is widely used to actively protect the Mission-Critical Systems. Although the MTD strategy has shifted the paradigm from passive system defense to active system defense, the indiscriminate use of the MTD strategy has the disadvantage of acting as a large overhead on the system to be protected. To solve this problem, in this paper, we derive the attack surface of the system to be protected using cyber attack information (OpenIOC). Then, based on the derived data, we propose a data visualization engine that can help configure a systematic MTD strategy by linking it with MTD strategy components. Through the proposed data visualization engine, existing and new MTD strategy researchers can configure a more systematic MTD strategy.

This survey explores electricity theft detection in smart grids, where traditional power systems meet modern technology. Smart grids, designed for efficient energy management and continuous integration of renewables, face a pressing challenge electricity theft, costing utility companies over $96 billion annually. The survey traces the evolution from conventional to smart grids, emphasizing their core components. It underscores the economic impact of theft, driving researchers to explore Artificial Intelligence (AI) and Deep Learning (DL) techniques for detection. A comprehensive literature review reveals various approaches, with a focus on DL's growing influence. Public datasets are explored as invaluable resources, and methods for theft detection, including advanced AI and DL, are dissected. Performance metrics like accuracy and precision are discussed, and challenges, including imbalanced data and privacy concerns, are highlighted. In conclusion, the survey emphasizes the need for diverse AI and DL approaches, data sources, and features to create robust theft detection systems for smart grids, ensuring their secure and efficient operation.

Rapid technological advancement and industrial growth pervade modern society. Light pollution has emerged as a significant environmental issue, resulting from excessive artificial nighttime lighting. It obscures celestial views and negatively impacts human health, contributing to sleep disorders and mood imbalances. This article introduces an immersive virtual reality (VR) based learning approach to reduce light pollution and raise awareness about the issue of light pollution. Utilized the META QUEST 2 Head-Mounted Display in a virtual environment, enable users to actively participate in learning scenarios. A variety of terrain textures, city models, furniture, and lighting effects are used to interact with miniatures in a virtual room, influencing the city's lighting and unveiling a starlit sky. The impact of the proposed work is evaluated through experiments and an object recognition module, showing a significant increase in awareness and understanding of light pollution and its effects. By providing an experiential learning opportunity, this tool has the potential to encourage proactive engagement, contribute to efforts to reduce light pollution and promote environmental well-being.

Weld defect inspection is essential to ensuring the safety of weld joints. However, this is a subjective, complex, and labor-intensive task for workers. To relieve this problem, this paper aims to weld defect detection tasks by applying the stateof- the-art YOLOv5x-seg by modifying the YOLOv5 network. In particular, we attempt to utilize the pixel-level polygon representation. Experimental results show that it achieves 82.6% mAP@0.5. In conclusion, our result shows that YOLOv5xseg can successfully perform weld defect detection tasks.

The problem of discrimination due to information alienation among Korean sign language users is continuously mentioned, and sign language translation research is actively being conducted to solve this problem. However, due to technological limits in translating text into Korean Sign Language, the need for specialist equipment causes annoyance and spatial constraints. Furthermore, it isn't easy to replicate the vocabulary and grammatical structure of the Korean language in Korean Sign Language. Furthermore, the service's commercialization is complicated by the need for more nonmanual signal (NMS) identification technology.

The commercialization of Internet of Things (IoT) technology has enabled data-driven connectivity and automation across a variety of industries. In particular, the manufacturing field has developed into a smart factory that automates parts of the production process and optimizes manufacturing activities through data collection and analysis. In this smart factory environment, as security and scalability issues regarding large amounts of data generated within the network arise, the need for distributed ledger-based solutions is emerging. In this paper, we design a Directed Acyclic Graph (DAG) based distributed data management system that enables distributed data processing in a smart factory IoT network composed of multiple devices. The proposed system uses a DAGbased data storage structure and state channel technology to improve network efficiency, security, and scalability. The proposed system can comprehensively support heterogeneous IoT devices and ensure the integrity of data transmitted and received in this process.

As global energy consumption continues to rise, artificial intelligence-based models for efficient energy usage are emerging worldwide. Traditional learning approaches necessitate extensive datasets to create accurate models for predicting energy consumption. However, acquiring the vast amount of data required by these models has its limitations. Therefore, we propose a model for accurate energy consumption prediction using transfer learning, allowing effective modeling with smaller datasets. In this study, we present a deep learning model for energy consumption prediction, utilizing transfer learning on various energy domain datasets. We construct a base model using CNN-Seq2Seq with input variables such as heat consumption and meteorological data. For comparative evaluation, we employ SVR, XGBoost, LightGBM, Random Forest, LSTM, and Seq2Seq models, utilizing metrics like MAE, MSE, and R2 Score. We explore the impact of applying transfer learning on source domain data to four distinct target domain datasets, comparing results with and without transfer learning, as well as with other machine learning algorithm models. The findings demonstrate that applying transfer learning yields superior accuracy across all four target domain datasets. This suggests the potential of overcoming limitations in specific energy domains and obtaining meaningful results through the application of transfer learning.

This paper analyzes underwater OFDM system's timing synchronization performance by simulating the channel impulse response of three Korean Peninsula coastal sea based on HFM and Zadoff-chu sequences (ZCS). Simulation results show that the timing synchronization performance varies depending on the characteristics of the three seas, and the ZCSbased fine synchronization algorithm improves the synchronization performance in all three sea areas compared to coarse synchronization.

As the importance of artificial intelligence (AI) as a diagnostic aid in the medical field is gradually increasing, our study constructed an AI model that detects root canal treatment areas using oral and maxillofacial data. We constructed models using three types (v5s, v5l, v5x) of the real-time object detection algorithm YOLO (You Only Look Once) version 5 to meet the medical field's requirement for more precise, faster, and accurate performance. Each model was trained for 300 epochs using an SGD optimizer and as a result of the experiment, all versions of YOLO v5 algorithms showed high mAP@.5 performance over 0.93. However, for mAP@ .5:.95 performance which corresponds to more precise detection performance evaluation, it was confirmed that there is a difference in performance depending on the network size of the model. Thus, we suggest that YOLO v5x model with the largest network size is most suitable for detecting root canal treatment areas. Through this research, we suggest future research directions in fields related to development of diagnostic aids based on AI and look forward to developing more advanced object detection algorithms.

Recent deep learning models perform well in image and natural language processing. However, in tabular data, there is a problem that good performance is not achieved due to data-level problems. Recently, TabNet, a model that overcomes these shortcomings, has been widely used for tabular data learning. However, categorical variable data does not perform significantly in tabular data. To solve this problem, Catboost Encoding method is used to solve the problem. In the case of this model, the pre-processing of categorical variable data was well utilized to derive more performance than other models, and it showed better performance than other encoding techniques.