Latest Issue
Vol. 52,  No. 9, Sep.  2025

All Issues

Buffer Management Algorithm Considering Characteristics of ZNS Interface

Gyupin Moon, Donghyun Kang

http://doi.org/10.5626/JOK.2025.52.9.721

Recently, there have been numerous efforts to enhance the performance and lifespan of SSDs by mapping their internal structure into zones. However, the write buffer algorithm within SSDs is still implemented in a traditional manner, which is unsuitable for SSDs with the Zoned Namespace (ZNS) interface. In this paper, we propose a new buffer management algorithm called Zone-aware Buffer (ZAB), which is designed to be compatible with the ZNS interface. ZAB divides the write buffer space into zones to minimize performance interference between them and to exploit parallelism within the SSD, thereby improving write performance. We implement ZAB based on ConfZNS, and our experiments demonstrate that ZAB improves performance by up to 4.6 times compared to systems that do not use a write buffer.

Software-Level GPU Preemption via OpenCL Kernel Scheduling in User Space

Namcheol Lee, Geonha Park, Woobean Seo, Seongsoo Hong

http://doi.org/10.5626/JOK.2025.52.9.727

As AI becomes increasingly prevalent in modern embedded systems, ensuring real-time performance for deep neural network (DNN) inference using GPUs has become a critical challenge. However, since the GPU is a shared resource with high preemption costs, process-level preemption is not effectively supported. As a result, priority inversion occurs during resource contention among processes, complicating the implementation of real-time multitasking in embedded systems utilizing GPUs. Previous studies have explored hardware-dependent approaches to GPU preemption, but these methods often lack portability and scalability. To overcome these limitations, this study proposes a software-level GPU preemption technique that enables preemption without relying on hardware-specific mechanisms. The proposed method intercepts GPU kernel execution requests from processes and forwards them to a user-space OpenCL kernel scheduler, which controls the execution order of GPU kernels based on process priorities. This approach reduces delays for high-priority processes caused by lower-priority ones. Experimental results confirm that the proposed method achieves high execution determinism.

SRGM-Based Reliability Evaluation of Cross-Domain Solutions

Eunjeong Ju, Jeonghwa Lee, Duksan Ryu

http://doi.org/10.5626/JOK.2025.52.9.738

With the accelerated development of Industrial Control Systems (ICS), critical cyber attacks continue to pose a persistent threat. Given the prevalence of ICS in nationally significant facilities, the consequences of these attacks are more severe compared to other domains. it is essential to proactively eliminate security vulnerabilities in ICS operational environments, particularly in bidirectional communication-capable devices used for control protocol traffic stability testing. Since vulnerabilities often arise from software errors, reducing these errors is crucial. Evaluating reliability, which informs testing frequency, becomes essential however,it is also time and resource-intensive. This study propose a reliability assessment approach for the testing stages of control protocol traffic stability testing equipment. It suggests employing a software reliability growth model and applying CODAS-E to identify an optimized software growth model tailored to the research objectives.The reliability assessment approach developed in this study is expected to enhance the quality of cross-domain systems.

Enhancing Sarcopenia Prediction with Genetic Algorithms for Feature Selection

Jiwoo Song, Jaehyeong Lee, Yourim Yoon

http://doi.org/10.5626/JOK.2025.52.9.749

Sarcopenia, a disease predominantly affecting the elderly, has emerged as a significant concern within the medical community. Due to the variety of sarcopenia's causes and diagnostic methods, identifying a specific cause remains challenging, which hampers the ability to predict it accurately with current knowledge. This study utilizes survey data from the Korean Longitudinal Study of Aging (KLoSA) to explore ways to enhance the accuracy of sarcopenia prediction through data preprocessing and feature selection using genetic algorithms. Data preprocessing reduced the number of features from 2,756 to 613. Subsequently, feature selection was performed and evaluated with logistic regression, XGBoost, and random forest as classification algorithms, achieving an accuracy of up to 84.73% and an F1-Score of 0.5953. These findings suggest practical insights into the effective application of genetic algorithms for analyzing survey-type data, potentially improving sarcopenia diagnosis.

Mixed Sound Source Localization via Audio-Visual Information Fusion

YuEun Lee, Sung Jin Um, Jung Uk Kim

http://doi.org/10.5626/JOK.2025.52.9.762

Multi-source localization is a research topic that uses audio mixed with multiple sources within a visual scene to identify the locations of individual sound sources. Existing studies have limitations in that they primarily use auditory information to assist the spatial domain of visual information, and they require prior knowledge information integration module that fuses audio-visual information, allowing auditory information to be utilized alongside spatial cues and visual information. Additionally, we introduce an object repetition detection module designed to identify objects that produce sounds repeatedly, enabling effective localization and separation of multiple sound sources without needing prior knowledge of the number of objects. The proposed method address the limitations of existing studies and enhances sound source localization capabilities. We also conducted experiments on the VGGSound dataset and achieved better performance than existing approaches.

Diffusion Model-based In-vehicle Noise Augmentation through Expert Knowledge-guided Clustering

Seok-Hun Choi, Mugeun Baek, Seok-Jun Buu

http://doi.org/10.5626/JOK.2025.52.9.771

To ensure vehicle operational safety and enhance user experience, it is crucial to accurately classify in-vehicle noise and detect performance anomalies in advance. However, deep learning-based noise classifiers often struggle in complex acoustic environments, such as those with external noise and internal reverberation. To address these challenges, we propose a novel vehicle noise classification method that integrates diffusion model-based signal augmentation with expert knowledge-guided clustering. This approach synthesizes a variety of challenging in-vehicle acoustic conditions and enhances signal-label associations through automatic label assignment based on expert-defined clusters. As a result, we can create training datasets that closely mirror real-world scenarios. Our experiments demonstrate that this method achieves a classification accuracy of 99.60%, surpassing state-of-the-art classifiers and improving by 0.06 percentage points over existing generative augmentation methods, thereby showcasing the effectiveness of the diffusion-based approach.

Resolving Ambiguity in Visual Question Answering through an Iterative Clarifying QA-based Framework

Yu-Jeong Sung, Gyu-Min Park, Seong-Bae Park

http://doi.org/10.5626/JOK.2025.52.9.778

This paper presents a three-stage framework to tackle the problem of ambiguous objects in Visual Question Answering (VQA), where the object referred to in a question is unclear due to multiple candidates in the image. The framework includes: (1) detecting whether the question is ambiguous, (2) generating clarification questions when ambiguity is detected, and (3) utilizing the Q&A history to perform the final VQA. Clarification questions are generated directly by the model, leveraging visual features without any additional training. The model iteratively refines its questions by incorporating the history of previous question-answer pairs. Experiments using the LLaVA v1.6 model demonstrate that the proposed framework enhances accuracy by 6.7% and semantic accuracy by 5.6% compared to the baseline. Moreover, the integration of ambiguity detection and an early stopping strategy reduces the inefficiencies associated with multi-turn interactions, resulting in a 44% decrease in execution time. This study offers a practical solution to the ambiguous objects problem by enabling real-time clarification without the need for additional training, ultimately leading to improved VQA accuracy.

Anomaly Detection in Mammography Data using Dimensionality Reduction and DBSCAN Clustering for Enhancing Diagnostic Model Performance

Donghee Kim, Jaeil Kim

http://doi.org/10.5626/JOK.2025.52.9.787

This study introduces a data cleaning technique for identifying and removing anomalous images from mammography data. An autoencoder extracts low-dimensional latent features, which are then refined through dimensionality reduction (methods such as PCA, t-SNE, and Isomap) to highlight irregular patterns. DBSCAN clustering is subsequently employed to detect anomalies. An ablation study confirmed that dimensionality reduction enhances anomaly detection, and the impact of anomaly removal on model training was assessed. Results indicate that the combination of t-SNE and DBSCAN yields superior performance, with the refined model demonstrating significant improvements in accuracy and sensitivity. These findings enhance the reliability of AI-based breast cancer diagnosis and present a promising pre-processing method for medical imaging.

Comparing RSSI Data Augmentation and Indoor Localization Techniques for Application in Manufacturing Environments

Hee-Jun Lee, Sang-Hwa Chung, Jeongbae Park

http://doi.org/10.5626/JOK.2025.52.9.795

The smartification of factories is progressing rapidly in many manufacturing environments. To digitalize maintenance processes, we have developed an innovative smart E-Ink tag system. A key feature of this system is its RSSI-based BLE smart tag localization function, which determines the real-time location of smart tags attached to process logistics. However, gathering sufficient RSSI fingerprint data to establish the system presents several challenges, including interference from operational equipment, workers, metal structures, and other obstacles, which can consume considerable time and resources. This study aims to enhance the limited RSSI fingerprint data collected from a bolt manufacturing factory and improve the localization performance of smart tags. We compare and analyze the errors and effectiveness of various data augmentation techniques alongside machine learning-based localization methods, including WkNN, Random Forest, Multilayer Perceptron, and LSTM.


Search




Journal of KIISE

  • ISSN : 2383-630X(Print)
  • ISSN : 2383-6296(Electronic)
  • KCI Accredited Journal

Editorial Office

  • Tel. +82-2-588-9240
  • Fax. +82-2-521-1352
  • E-mail. chwoo@kiise.or.kr