• KNOM Review
• /
• v.23 no.2
• /
• pp.18-28
• /
• 2020

With the development of the virtual community, the benefits that IT technology provides to people in fields such as healthcare, industry, communication, and culture are increasing, and the quality of life is also improving. Accordingly, there are various malicious attacks targeting the developed network environment. Firewalls and intrusion detection systems exist to detect these attacks in advance, but there is a limit to detecting malicious attacks that are evolving day by day. In order to solve this problem, intrusion detection research using machine learning is being actively conducted, but false positives and false negatives are occurring due to imbalance of the learning dataset. In this paper, a Random Oversampling method is used to solve the unbalance problem of the UNSW-NB15 dataset used for network intrusion detection. And through experiments, we compared and analyzed the accuracy, precision, recall, F1-score, training and prediction time, and hardware resource consumption of the models. Based on this study using the Random Oversampling method, we develop a more efficient network intrusion detection model study using other methods and high-performance models that can solve the unbalanced data problem.

• KIPS Transactions on Computer and Communication Systems
• /
• v.12 no.8
• /
• pp.253-262
• /
• 2023

E-voting is a concept that includes actions such as kiosk voting at a designated place and internet voting at an unspecified place, and has emerged to alleviate the problem of consuming a lot of resources and costs when conducting offline voting. Using E-voting has many advantages over existing voting systems, such as increased efficiency in voting and ballot counting, reduced costs, increased voting rate, and reduced errors. However, centralized E-voting has not received attention in public elections and voting on corporate agendas because the results of voting cannot be trusted due to concerns about data forgery and modulation and hacking by others. In order to solve this problem, recently, by designing an E-voting system using blockchain, research has been actively conducted to supplement concepts lacking in existing E-voting, such as increasing the reliability of voting information and securing transparency. In this paper, we proposed an electronic voting system that introduced hybrid blockchain that uses public and private blockchains in convergence. A hybrid blockchain can solve the problem of slow transaction processing speed, expensive fee by using a private blockchain, and can supplement for the lack of transparency and data integrity of transactions through a public blockchain. In addition, the proposed system is implemented as BaaS to ensure the ease of type conversion and scalability of blockchain and to provide powerful computing power. BaaS is an abbreviation of Blockchain as a Service, which is one of the cloud computing technologies and means a service that provides a blockchain platform ans software through the internet. In this paper, in order to evaluate the feasibility, the proposed system and domestic and foreign electronic voting-related studies are compared and analyzed in terms of blockchain type, anonymity, verification process, smart contract, performance, and scalability.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.33 no.2
• /
• pp.267-280
• /
• 2023

With development of computing and communications technologies, IoT environments based on high-speed networks have been extending rapidly. Especially, from home to an office or a factory, applications of IoT devices with sensing environment and performing computations are increasing. Unfortunately, IoT devices which have limited hardware resources can be vulnerable to cyber attacks. Hence, there is a concern that an IoT botnet can give rise to information leakage as a national cyber security crisis arising from abuse as a malicious waypoint or propagation through connected networks. In order to response in advance from unknown cyber threats in IoT networks, in this paper, We firstly define four types of We firstly define four types of characteristics by analyzing darknet traffic accessed from an IoT botnet. Using the characteristic, a suspicious IP address is filtered quickly. Secondly, the filtered address is identified by Cyber Threat Intelligence (CTI) or Open Source INTelligence (OSINT) in terms of an unknown suspicious host. The identified IP address is finally fingerprinted to determine whether the IP is a malicious host or not. To verify a validation of the proposed method, we apply to a Darknet on real-world SOC. As a result, about 1,000 hosts who are detected and blocked preemptively by the proposed method are confirmed as real IoT botnets.

• Journal of the Korean GEO-environmental Society
• /
• v.25 no.4
• /
• pp.13-20
• /
• 2024

In this paper, a parallel analysis algorithm for Smoothed Particle Hydrodynamics (SPH), one of the numerical methods for fluidic materials, is introduced. SPH, which is a meshless method, can represent the behavior of a continuum using a particle-based approach, but it demands substantial computational resources. Therefore, parallel analysis algorithms are essential for SPH simulations. The domain decomposition algorithm, which divides the computational domain into partitions to be independently analyzed, is the most representative method among parallel analysis algorithms. In Discrete Element Method (DEM) and Molecular Dynamics (MD), the Cartesian coordinate-based domain decomposition method is popularly used because it offers advantages in quickly and conveniently accessing particle positions. However, in SPH, it is important to share particle information among partitioned domains because SPH particles are defined based on information from nearby particles within the smoothing length. Additionally, maintaining CPU load balance is crucial. In this study, a highly parallel efficient algorithm is proposed to dynamically minimize the size of orthogonal domain partitions to prevent excess CPU utilization. The efficiency of the proposed method was validated through numerical analysis models. The parallel efficiency of the proposed method is evaluated for up to 30 CPUs for fluidic models, achieving 90% parallel efficiency for up to 28 physical cores.

• Journal of the Korea Computer Graphics Society
• /
• v.30 no.3
• /
• pp.99-107
• /
• 2024

While the demand for extrapolating video content horizontally or vertically is increasing, even the most advanced techniques cannot successfully extrapolate all videos. Therefore, it is important to determine if a given video can be well extrapolated before attempting the actual extrapolation. This can help avoid wasting computing resources. This paper proposes a video classifier that can identify if a video is suitable for horizontal extrapolation. The classifier utilizes optical flow and an adaptive Gaussian blur network, which can be applied to flow-based video extrapolation methods. The labeling for training was rigorously conducted through user tests and quantitative evaluations. As a result of learning from this labeled dataset, a network was developed to determine the extrapolation capability of a given video. The proposed classifier achieved much more accurate classification performance than methods that simply use the original video or fixed blur alone by effectively capturing the characteristics of the video through optical flow and adaptive Gaussian blur network. This classifier can be utilized in various fields in conjunction with automatic video extrapolation techniques for immersive viewing experiences.

• Proceedings of the Korea Contents Association Conference
• /
• 2007.11a
• /
• pp.560-564
• /
• 2007

Korea Basic Science Institute(KBSI) has three general electron microscopes including High Voltage Electron Microscope(HVEM) which is the only one in Korea. Observed images through an electron microscope are what they are tilted by each step and saved, offering the more better circumstances for observers, a reconstruction to 3D could be a essential process. In this process, a warping method decreases distortions maximumly of avoided parts of a camera's focus. All these image treatment processes and 3D reconstruction processes are based on an accompaniment of a highly efficient computer, a number of Grid Node Personal computers share this process in a short time and dispose of it. Grid Node Personal computers' purpose is to make an owner can share different each other and various computing resources efficiently and also Grid Node Personal computers is applying to solve problems like a role scheduling needed for a constructing system, a resource management, a security, a capacity measurement, a condition monitoring and so on. Grid Node Personal computers accomplish roles of a highly efficient computer that general individuals felt hard to use, moreover, a image treatment using the warping method becomes a foundation for reconstructing to more closer shape with an real object of observation. Construction of the electron microscope volume 랜더링 system based on Grid Node Personal computer through the warping process can offer more convenient and speedy experiment circumstances to observers, and makes them meet with experiment outcome that is similar to real shapes and is easy to understand.

• Journal of Intelligence and Information Systems
• /
• v.28 no.4
• /
• pp.251-273
• /
• 2022

Recently, research on applying text analysis to deep learning has steadily continued. In particular, researches have been actively conducted to understand the meaning of words and perform tasks such as summarization and sentiment classification through a pre-trained language model that learns large datasets. However, existing pre-trained language models show limitations in that they do not understand specific domains well. Therefore, in recent years, the flow of research has shifted toward creating a language model specialized for a particular domain. Domain-specific pre-trained language models allow the model to understand the knowledge of a particular domain better and reveal performance improvements on various tasks in the field. However, domain-specific further pre-training is expensive to acquire corpus data of the target domain. Furthermore, many cases have reported that performance improvement after further pre-training is insignificant in some domains. As such, it is difficult to decide to develop a domain-specific pre-trained language model, while it is not clear whether the performance will be improved dramatically. In this paper, we present a way to proactively check the expected performance improvement by further pre-training in a domain before actually performing further pre-training. Specifically, after selecting three domains, we measured the increase in classification accuracy through further pre-training in each domain. We also developed and presented new indicators to estimate the specificity of the domain based on the normalized frequency of the keywords used in each domain. Finally, we conducted classification using a pre-trained language model and a domain-specific pre-trained language model of three domains. As a result, we confirmed that the higher the domain specificity index, the higher the performance improvement through further pre-training.

• Journal of the Korean earth science society
• /
• v.43 no.6
• /
• pp.691-711
• /
• 2022

The state-of-the-art Earth system model as a virtual Earth is required for studies of current and future climate change or climate crises. This complex numerical model can account for almost all human activities and natural phenomena affecting the atmosphere of Earth. The Unified Model (UM) from the United Kingdom Meteorological Office (UK Met Office) is among the best Earth system models as a scientific tool for studying the atmosphere. However, owing to the expansive numerical integration cost and substantial output size required to maintain the UM, individual research groups have had to rely only on supercomputers. The limitations of computer resources, especially the computer environment being blocked from outside network connections, reduce the efficiency and effectiveness of conducting research using the model, as well as improving the component codes. Therefore, this study has presented detailed guidance for installing a new version of the UM on high-performance parallel computers (Linux clusters) owned by individual researchers, which would help researchers to easily work with the UM. The numerical integration performance of the UM on Linux clusters was also evaluated for two different model resolutions, namely N96L85 (1.875° ×1.25° with 85 vertical levels up to 85 km) and N48L70 (3.75° ×2.5° with 70 vertical levels up to 80 km). The one-month integration times using 256 cores for the AMIP and CMIP simulations of N96L85 resolution were 169 and 205 min, respectively. The one-month integration time for an N48L70 AMIP run using 252 cores was 33 min. Simulated results on 2-m surface temperature and precipitation intensity were compared with ERA5 re-analysis data. The spatial distributions of the simulated results were qualitatively compared to those of ERA5 in terms of spatial distribution, despite the quantitative differences caused by different resolutions and atmosphere-ocean coupling. In conclusion, this study has confirmed that UM can be successfully installed and used in high-performance Linux clusters.

• Journal of Intelligence and Information Systems
• /
• v.24 no.1
• /
• pp.205-225
• /
• 2018

Convolutional Neural Network (ConvNet) is one class of the powerful Deep Neural Network that can analyze and learn hierarchies of visual features. Originally, first neural network (Neocognitron) was introduced in the 80s. At that time, the neural network was not broadly used in both industry and academic field by cause of large-scale dataset shortage and low computational power. However, after a few decades later in 2012, Krizhevsky made a breakthrough on ILSVRC-12 visual recognition competition using Convolutional Neural Network. That breakthrough revived people interest in the neural network. The success of Convolutional Neural Network is achieved with two main factors. First of them is the emergence of advanced hardware (GPUs) for sufficient parallel computation. Second is the availability of large-scale datasets such as ImageNet (ILSVRC) dataset for training. Unfortunately, many new domains are bottlenecked by these factors. For most domains, it is difficult and requires lots of effort to gather large-scale dataset to train a ConvNet. Moreover, even if we have a large-scale dataset, training ConvNet from scratch is required expensive resource and time-consuming. These two obstacles can be solved by using transfer learning. Transfer learning is a method for transferring the knowledge from a source domain to new domain. There are two major Transfer learning cases. First one is ConvNet as fixed feature extractor, and the second one is Fine-tune the ConvNet on a new dataset. In the first case, using pre-trained ConvNet (such as on ImageNet) to compute feed-forward activations of the image into the ConvNet and extract activation features from specific layers. In the second case, replacing and retraining the ConvNet classifier on the new dataset, then fine-tune the weights of the pre-trained network with the backpropagation. In this paper, we focus on using multiple ConvNet layers as a fixed feature extractor only. However, applying features with high dimensional complexity that is directly extracted from multiple ConvNet layers is still a challenging problem. We observe that features extracted from multiple ConvNet layers address the different characteristics of the image which means better representation could be obtained by finding the optimal combination of multiple ConvNet layers. Based on that observation, we propose to employ multiple ConvNet layer representations for transfer learning instead of a single ConvNet layer representation. Overall, our primary pipeline has three steps. Firstly, images from target task are given as input to ConvNet, then that image will be feed-forwarded into pre-trained AlexNet, and the activation features from three fully connected convolutional layers are extracted. Secondly, activation features of three ConvNet layers are concatenated to obtain multiple ConvNet layers representation because it will gain more information about an image. When three fully connected layer features concatenated, the occurring image representation would have 9192 (4096+4096+1000) dimension features. However, features extracted from multiple ConvNet layers are redundant and noisy since they are extracted from the same ConvNet. Thus, a third step, we will use Principal Component Analysis (PCA) to select salient features before the training phase. When salient features are obtained, the classifier can classify image more accurately, and the performance of transfer learning can be improved. To evaluate proposed method, experiments are conducted in three standard datasets (Caltech-256, VOC07, and SUN397) to compare multiple ConvNet layer representations against single ConvNet layer representation by using PCA for feature selection and dimension reduction. Our experiments demonstrated the importance of feature selection for multiple ConvNet layer representation. Moreover, our proposed approach achieved 75.6% accuracy compared to 73.9% accuracy achieved by FC7 layer on the Caltech-256 dataset, 73.1% accuracy compared to 69.2% accuracy achieved by FC8 layer on the VOC07 dataset, 52.2% accuracy compared to 48.7% accuracy achieved by FC7 layer on the SUN397 dataset. We also showed that our proposed approach achieved superior performance, 2.8%, 2.1% and 3.1% accuracy improvement on Caltech-256, VOC07, and SUN397 dataset respectively compare to existing work.