Earticle

There have been active researches on the prediction of future events with moving (first person view) cameras by using deep-learning approaches. In many those approaches, the moving camera motion was considered as an important cue in future object localization and anticipation of traffic accidents. This paper argues that the moving cameramotion (i.e., ego-motion) is not necessary for these tasks because only the relative motion of the surrounding objects with respect to the moving camera is sufficient for those tasks. In this paper, we will present some empirical evidence from the recently published papers with codes and datasets. Comparison results with and without camera-motion shows that performance differences are rather minor. This may recommend not to use the camera-motion for reducing memories as well as inference time at edge devices because the number of parameters in the architecture can be significantly reduced.

Artificial intelligence (AI) technology has been adapted to innovate the display industry. Accordingly, display companies need to train their empolyees to have basic understanding for AI. Over the past three years, we have conducted AI education for general empolyees of companies in the display-related fields. In this paper, we introduce the developments of AI curriculums and analyze their effects on satififaction of trainees. The result show that curriculum with more online education provides more satisfaction and job application of employees.

Medical images constitute a substantial portion of all medical data, but various issues arise, including noise and judgment-related problems. Therefore, recent research has actively explored deep learning applications, such as noise removal and disease classification based on medical images. In particular, brain tumors are typically diagnosed using MRI, and early detection is crucial. In this study, the classification of MRI images for brain tumor patients was conducted by improving MRI noise and utilizing a pre-trained CNN model.

This study aims to select the factors affecting depression in young people and to identify the relationship by using a hybrid machine learning. From 2018 to 2022, a total of four years of community health survey data were used, and the final subjects of the study were 141,510 young people aged 19 to 34. In order to identify the factors that significantly affect depression, a feature selection with Least absolute shrinkage and selection operator (LASSO) regression has been applied. After selecting variables, logistic regression analysis was performed through a complex sampling design. Through feature selection, variables such as health behavior (smoking, drinking, sleeping) and health condition (with or without hypertension) were selected. Among the variables related to health behavior, smoking (OR 1.95; 95% CI 1.82 - 2.09) and sleep_7 hours or more (OR 0.70; 95% CI 0.66 - 0.74) had a significant effect on depression. Both the hypertension and subjective health level, which are variables related to health status, had a significant effect on depression.

LCA(Life Cycle Assessment) has begun drawing attention because of EU’s carbon border tax and it is commonly used to asses environmental impacts quantitatively. This research aims to design Korea-typed LCA (K-LCA) software architecture as an effective software tool that meets criteria of ISO 14000 series. The proposed K-LCA software has been designed to provide Flexibility and Integration, High Granularity LCA Data, Unified Description Format, Modularity and Simulation Expansion, Intuitive User Interface based on LCA methodology. Also it has been designed to provide a novel data-driven LCA interpretation assistant scheme using graph theory for enhancing the software performance and simulation capabilities. It contributes to enable quantitative environmental impact assessment ensuring precise and accurate LCA results.

Neural networks are increasingly being used in cloud-based applications, which require users to upload their sensitive data to the cloud server. However, the data privacy may be compromised when the server trains or infers a neural network model using the plaintext data. To address this privacy issue, many studies have developed privacy-preserving neural networks. Recently, FENet, a privacy-preserving neural networks framework using functional encryption, was proposed by Panzade and Takabi. In this paper, we propose a method to accelerate the secure activation function of FENet. We adopt a precomputation approach to reduce the computational overhead of privacy-preserving matrix multiplication, which is the dominant operation in the secure activation function of FENet. According to our performance analysis, the privacypreserving matrix multiplication can be performed by 3.77 times faster than that of FENet with additional 3.49 MB of memory. Since the secure activation function of FENet can be applied to both the training and inference phases, the proposed method is expected to accelerate both phases.

Acute myocardial infarction (AMI) is one of the leading causes of cardiovascular disease-related mortality worldwide, causing ischemic damage to the heart muscle. AMI poses a risk of sudden cardiac arrest and is associated with a high rate of recurrence, which can significantly impact daily life. Therefore, this study aims to develop a predictive model for mortality in AMI patients using machine learning. The data used in the study are from KNHDIS (2013-2022) and include demographic characteristics and disease information of discharged patients. The model was constructed using RFECV and Random Forest. Key variables influencing mortality include age, number of surgeries, and length of hospital stay. The model demonstrated high performance with an accuracy of 94% and an AUC value of 0.93.

Sustainable power systems should include solar energy generation. However, for effective grid management and the integration of renewable energy sources, accurate solar power generation predictions are essential. Therefore, this study compares the prediction of solar power forecasting in Italy and Bulgaria. These are two countries that have alike latitudes but different populations and solar energy production. The historical solar power generation and meteorological data from these countries are preprocessed and then used to apply four different deep learning models including Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). The results are analyzed to gain insights into how the proximity of geographical locations and the quality and quantity of data impact the precision of prediction algorithms.

In recent years, Generative Adversarial Networks (GANs) appeared as a prevailing solution for combating data scarcity in various domains. This study delves into utilizing WaveGAN, a specialized GAN architecture, to address the inherent challenges stemming from the limited availability of audio datasets. Our primary objective is to tackle the issue of constrained audio data resources by utilizing the potential of WaveGAN. Our research is driven by the overarching goal of investigating the capacity of CNN to gather significant insights from an extensive corpus of human speech data. A key focus of our work is to demonstrate the effectiveness of WaveGAN in generating synthetic audio data, thereby expanding the breadth of our audio dataset and bolstering the resilience of our classification models. Our study aims to yield improved classification results, providing crucial insights into the viability of this approach in alleviating data scarcity challenges of audio analysis.

Cyberattacks have affected every technology in recent years, and the metaverse is not an exception. Users can engage with other users by navigating the metaverse with their avatar's associated user data. Attackers using the user data associated with avatars masquerade as users or penetrate through the metaverse's vulnerable infrastructure. There are multiple components of the metaverse infrastructure that are vulnerable to attack. The infrastructure becomes insecure when these components are developed without security in consideration. Cyberattacks that take advantage of these vulnerabilities compromise user privacy and cause the loss of sensitive data. This highlights the necessity of comprehending the infrastructure and the data flow through its essential components so that appropriate countermeasures can be built. Our study proposed a model that clarifies the data flow among the infrastructure's components for creating an avatar in the metaverse. We also identify the major threat causing cyberattacks per component. We state in our study's conclusion that such threats will have an impact on the metaverse's avatar and user data.

This research proposes the development of an online psychological counseling platform utilizing an Artificial Intelligence (AI)-based emotion analysis system. The platform, leveraging facial video, voice, and text data, aims to real-time identify and analyze the emotional states of counselees in a non-face-to-face counseling environment, providing counselors with the necessary information to facilitate appropriate counseling. The outcomes of this research are expected to enhance communication effectiveness between counselors and counselees and contribute to the psychological well-being of counselees in online counseling scenarios.

This study emphasizes the necessity of artificial intelligence for rapid and accurate battery state-of-charge (SOC) prediction, a critical parameter in battery condition prediction. We compared and evaluated time series models previously used for SOC prediction, namely LSTM, GRU, and Transformer. In addition to model comparison, we experimented with data preprocessing techniques suitable for battery SOC prediction. The study utilized NASA's aging dataset comprising different cells under various experimental conditions. A Sliding Window technique was employed to multiply data and evaluate model performance. The results showed that the GRU model most effectively predicted battery SOC without data multiplication. However, after applying the Sliding Window technique to generate more learning data, the Transformer model outperformed others with an average RMSE of 0.032 and MAE of 0.006 across all batteries. This research paves the way for advancements in AI technology based on Transformer models for improved analysis of battery conditions, which can benefit manufacturing and recycling processes.

Respiratory diseases are one of the major causes of death worldwide. Therefore, research on respiratory disease classification using respiratory data is considered an important task. Previous studies mainly focused on respiratory disease classification using 2D feature extraction methods such as spectrograms and MFCCs. However, these methods have drawbacks such as long classification time and decreased accuracy as the number of respiratory disease types increases. To address this issue, we propose a solution that combines data with different dimensions to improve the performance of respiratory disease classification. We utilize the gammatone based spectrogram feature extraction method along with raw 1D respiratory data. By combining these two approaches, we can achieve both fast classification speed from 1D time-series models and high classification accuracy from 2D feature extraction methods. Our proposed respiratory disease classification study consists of four stages: data preprocessing, combined data generation, construction of a respiratory disease classification model, and decision-making for respiratory disease diagnosis. We validate our approach using a TCN (Temporal Convolutional Network) model and achieve a high respiratory disease classification accuracy of 98.93%. Moreover, our proposed method significantly reduces the training time for classification by more than four times compared to previous methods, thus demonstrating its superiority.

As deep learning technology has improved in recent years, it has expanded from text-oriented document analysis to unstructured data such as images and tables, and there are an increasing number of studies on extracting meaning from such data and analyzing documents. Among them, there are various studies that analyze chart images because charts provide a lot of information such as checking or comparing abnormal elements of data by graphically representing various types of data. Chart classification is an important step because each category has a different way of extracting data and extracting and interpreting the meaning accordingly, so the focus is on improving the classification performance of deep learning-based classification models for various chart categories. However, deep learning-based models have the problem that experts need to allocate a lot of time to configure the model design optimized for the data and check the performance of the model after training. As a way to alleviate these problems, in this paper, we studied a chart classification model using a Neural Architecture Search technique that automatically explores the model structure optimized for the data. The optimal network structure was explored, trained, and tested using 69,600 chart data consisting of 12 chart categories, and the performance was compared with chart classification models using VGG-16 and ResNet-50 algorithms as a way to check the performance of the model. The average classification performance of the model using Neural Architecture Search showed higher classification accuracy than other models with Precision 99.7%, Recall 99.6%, and F1-Score 99.9%.

Containers are one of the essential technologies for cloud services. Containers are lightweight and faster than traditional virtual machines as they directly utilize resources from the host. However, they suffer from drawbacks such as weaker security and isolation. To address these issues, security-focused runtimes like Kata, gVisor, and Firecracker have emerged. Each runtime has different creation methods, resulting in varying performance and resource usage. In this study, performance evaluations of these low-level runtimes were conducted.

Vehicular Adhoc Network (VANET) are vital in enhancing communication and safety in intelligent transportation systems, autonomous vehicles, and cooperative driving. Efficient and reliable data transmission in V2V scenarios is crucial for road safety and traffic management. However, high mobility and the dynamic nature of vehicular environments lead to challenges such as high packet loss, and packet queue length to mitigate network performance. To address these issues, we propose an innovative approach that leverages packet transmission data using weighted attention mechanism to predict packet flow and reduce congestion. We obtained a comprehensive dataset by integrating the H2V (Highway-to-Vehicle) based DSRC (Dedicated Short- Range Communication) protocol with the OMNET++ and SUMO simulators, encompassing network metrics such as queue length, packet loss and packet delay. The proposed LSTM approach shows promising results in predicting congestion patterns, enhancing the packet delivery ratio. Conclusively, our WDL-H2V model maintains a consistently high PDR to continuously mitigate congestion.

The important goal of the recent digital transformation paradigm is not the digitization of analog data, but the creation of new value through digital transformation [1]. In various industrial fields, technologies are being developed and applied to increase productivity and prevent safety accidents through ICT-based digital transformation. Companies in the domestic construction industry are trying to increase their global competitiveness through digital transformation of the entire construction industry cycle (design – construction – maintenance) by introducing smart construction technology. This paper introduces efficient underground facility maintenance technology and method based on smart construction technology.

The advancement of IT technology has led to an exponential increase in the demand for data processing. To address these challenges, businesses are utilizing multi-hybrid cloud architectures and a cloud-native approach to efficiently process and manage data. In this paper, we explores the cloud strategies adopted by contemporary businesses, particularly focusing on hybrid and multi-cloud technologies, and the cloudnative environment.

Smart-pipe is very important system for implementation of smart city. Because It will be checking status of various pipe. Through this, we can predict pipe defects in advance and prevent accidents. Therefore, we are conducting research on smart pipe management and defect prediction. In particular, in this study, we implemented visualization that managers can easily check to manage the status of smart pipes.

Planthopper is a major problem pest of agricultural crops and rice, feeding mainly on the leaves and stems of crops and causing devastating damage to farmers in South Korea. Planthoppers reduce nutrients in the body of crops and cause crop diseases by destroying tissue or transmitting viruses, so accurate detection and diagnosis is essential to minimize the damage. With the development of artificial intelligence in recent years, deep learning has been widely used to diagnose the pests. Most of the pest diagnosis research and programs use approaches based on object detection and image classification. However, traditional planthopper detection models may misidentify other pests as planthoppers, which can reduce user confidence in the diagnostic model. To address this misrecognition problem, this study investigates a deep learning-based planthopper Image detection and discrimination model for detecting and classifying the planthopper images. The proposed model combines the Faster RCNN object detection model and the Resnet50 classification model to automatically detect planthoppers among other pests in aerial entomology net images. The performance measurements showed that the benchmark model using the Faster RCNN algorithm achieved a high Recall of 91.23%, but a relatively low Precision of 24.34%. On the other hand, the model proposed in this study has a high recall of 96.22% and a high precision of 96.73%, proving that it can detect planthoppers well.