Dong-Jin Shin

Mr. Dong-Jin Shin

Department of Computer Engineering, Anyang University, Anyang-si, Republic of Korea
orcid

Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
7 April 2022 in Applied Sciences
Verified
Journal Article
Open access
PDF file
DSDong-Jin Shin
Dong-Jin Shin
KJKim Jeong-joon
Kim Jeong-joon
Deep learning-based object detection technology can efficiently infer results by utilizing graphics processing units (GPU). However, when using general deep learning frameworks in embedded systems and mobile devices, processing functionality is limited. This allows deep learning frameworks such as TensorFlow-Lite (TF-Lite) and TensorRT (TRT) to be optimized for different hardware. Therefore, this paper introduces a performance inference method that fuses the Jetson monitoring tool with TensorFlow and TRT source code on the Nvidia Jetson AGX Xavier platform. In addition, central processing unit (CPU) utilization, GPU utilization, object accuracy, latency, and power consumption of the deep learning framework were compared and analyzed. The model is You Look Only Once Version4 (YOLOv4), and the dataset uses Common Objects in Context (COCO) and PASCAL Visual Object Classes (VOC). We confirmed that using TensorFlow results in high latency. We also confirmed that TensorFlow-TensorRT (TF-TRT) and TRT using Tensor Cores provide the most efficiency. However, it was confirmed that TF-Lite showed the lowest performance because it utilizes a GPU limited to mobile devices. Through this paper, we think that when developing deep learning-related object detection technology on the Nvidia Jetson platform or desktop environment, services and research can be efficiently conducted through measurement results.

25 March 2022 in Symmetry
Verified
Journal Article
Open access
PDF file
SHSeung-Yeon Hwang
Seung-Yeon Hwang
DSDong-Jin Shin
Dong-Jin Shin
KJKim Jeong-joon
Kim Jeong-joon
Recently, as core technologies leading the fourth industrial revolution, such as the Internet of Things (IoT), 5G, the cloud, and big data, have promoted smart convergence across national socio-economic infrastructures, cyber systems are expanding and becoming complex, and they are not effective in responding to cyber safety risks and threats using security technology solutions limited to a single system. Therefore, we developed cyber security technology that combines machine learning and AI technology to solve complex problems related to cyber safety. In this regard, this study aims to identify technology development trends to prevent the risks and threats of various cyber systems by monitoring major cyber security convergence fields and technologies through the symmetrical thesis and patent analysis. Because thesis information can explain the superiority of technology and patent information can explain the usefulness of a technology, they can be effectively used for analyzing and predicting technology development trends. Therefore, in this study, latent Dirichlet allocation is applied to extract text-document-based technical topics for the symmetrical thesis and patent information to identify security convergence fields and technologies for cyber safety. In addition, it elucidates cyber security convergence fields and technology trends by applying a dynamic topic model and long short-term memory, which are useful for analyzing technological changes and predicting trends. Based on these results, cyber security administrators, system operators, and developers can effectively identify and respond to trends in related technologies to reduce threats, and companies and experts developing cyber security solutions can present a new security approach.

18 September 2021 in Applied Sciences
Verified
Journal Article
Open access
PDF file
DSDong-Jin Shin
Dong-Jin Shin
KJKim Jeong-joon
Kim Jeong-joon
Research has been conducted to efficiently transfer blocks and reduce network costs when decoding and recovering data from an erasure coding-based distributed file system. Technologies using software-defined network (SDN) controllers can collect and more efficiently manage network data. However, the bandwidth depends dynamically on the number of data transmitted on the network, and the data transfer time is inefficient owing to the longer latency of existing routing paths when nodes and switches fail. We propose deep Q-network erasure coding (DQN-EC) to solve routing problems by converging erasure coding with DQN to learn dynamically changing network elements. Using the SDN controller, DQN-EC collects the status, number, and block size of nodes possessing stored blocks during erasure coding. The fat-tree network topology used for experimental evaluation collects elements of typical network packets, the bandwidth of the nodes and switches, and other information. The data collected undergo deep reinforcement learning to avoid node and switch failures and provide optimized routing paths by selecting switches that efficiently conduct block transfers. DQN-EC achieves a 2.5-times-faster block transmission time and 0.4-times-higher network throughput than open shortest path first (OSPF) routing algorithms. The bottleneck bandwidth and transmission link cost can be reduced, improving the recovery time approximately twofold.

6 April 2023 in Symmetry
Verified
Journal Article
Open access
PDF file
DSDong-Jin Shin
Dong-Jin Shin
KJKim Jeong-joon
Kim Jeong-joon
With the development of various information and communication technologies, the amount of big data has increased, and distributed file systems have emerged to store them stably. The replication technique divides the original data into blocks and writes them on multiple servers for redundancy and fault tolerance. However, there is a symmetrical space efficiency problem that arises from the need to store blocks larger than the original data. When storing data, the Erasure Coding (EC) technique generates parity blocks through encoding calculations and writes them separately on each server for fault tolerance and data recovery purposes. Even if a specific server fails, original data can still be recovered through decoding calculations using the parity blocks stored on the remaining servers. However, matrices generated during encoding and decoding are redundantly generated during data writing and recovery, which leads to unnecessary overhead in distributed file systems. This paper proposes a cache-based matrix technique that uploads the matrices generated during encoding and decoding to cache memory and reuses them, rather than generating new matrices each time encoding or decoding occurs. The design of the cache memory applies the Weighting Size and Cost Replacement Policy (WSCRP) algorithm to efficiently upload and reuse matrices to cache memory using parameters known as weights and costs. Furthermore, the cache memory table can be managed efficiently because the weight–cost model sorts and updates matrices using specific parameters, which reduces replacement cost. The experiment utilized the Hadoop Distributed File System (HDFS) as the distributed file system, and the EC volume was composed of Reed–Solomon code with parameters (6, 3). As a result of the experiment, it was possible to reduce the write, read, and recovery times associated with encoding and decoding. In particular, for up to three node failures, systems using WSCRP were able to reduce recovery time by about 30 s compared to regular HDFS systems.