eISSN : 3022-7011
ISSUER : KIPS
 
After the Korea Information Processing Society (KIPS) Transactions journal was founded in 1994, it was reorganized into the KIPS Transactions: Computer and Communication Systems(2287-5891/2734-049X ) and the KIPS Transactions: Software and Data Engi neering(2287-5905/2734-0503) in 2012. Through the KIPS official meeting on January 8th, 2024, the new KIPS Transaction journal was founded by integrating two KIPS Journals, KIPS Transactions: Computer and Communication Systems and KIPS Transactions: Software and Data Engineering. The new journal aims to realize social value and contribute to the development of South Korea’s science and technology with support from the lottery fund of the Ministry of Strategy and Finance and the science/technology promotion fund of the Ministry of Science and ICT. It is indexed in the Korea Science Academic Database, Korea Citation Index (KCI), and EBSCO.

HighlightsMore

Diffusion-based Audio-to-Visual Generation for High-Quality Bird Images

Adel Toleubekova  Joo Yong Shim  XinYu Piao  Jong-Kook Kim
Accurately identifying bird species from their vocalizations and generating corresponding bird images is still a challenging task due to limited training data and environmental noise in audio data. To address this limitation, this paper introduces a...

Bambda: A Framework for Preventing Function Invocation Condition-Based Attacks in Serverless Environments

Shin Chang Hee  Lee Seung Soo
Serverless computing is rapidly emerging as a new paradigm in cloud computing, offering automatic scalability, cost efficiency, and ease of operation. However, its two core characteristics—IAM-based privilege management and event-driven execution—ca...

PEFT Methods for Domain Adaptation

Lee You Jin  Yoon Kyung Koo  Chung Woo Dam
This study analyzed that the biggest obstacle in deploying Large Language Models (LLMs) in industrial settings is incorporating domain specificity into the models. To mitigate this issue, the study compared model performance when domain knowledge wa...

Label Differential Privacy Study for Privacy Protection in Multimodal Contrastive Learning Model

Youngseo Kim  Minseo Yu  Younghan Lee  Ho Bae
Recent advancements in multimodal deep learning have garnered significant attention from both academia and industry due to their exceptional accuracy and ability to learn rich knowledge representations. In particular, contrastive learning based appr...

Latest Publication   (Vol. 14, No. 11, Nov.  2025)

Noise Attenuation and Precision Measurement Methods for Sensors Exposed to Electromagnetic Interference
Shin Hyun Jong  Jin-Goo Choi
This study proposes both hardware and software approaches to address the noise issues in sensors exposed to external environments. From the H/W perspective, a 500 Hz sine wave signal was used to measure the current of a resistive sensor, and a three-stage band-pass filter was implemented to effectively attenuate low- and high-frequency noise. On the S/W side, a microprocessor was utilized to generate a 500 Hz signal, and a median algorithm was applied to eliminate noise. The H/W approach achieved a noise attenuation effect of 90–95%, while the S/W approach effectively removed the remaining residual noise. These results confirm that an improved monitoring system with enhanced accuracy and precision can be realized.
The Transactions of the Korea Information Processing Society, Vol. 14, No. 11, pp. 855-864, Nov. 2025
https://doi.org/10.3745/TKIPS.2025.14.11.855
Noise attenuation Real-time Monitoring Bandpass filter Sine Wave Signal Median filtering
Stream Object Exchange Mechanism based on FHIR Server
Soyeon Kim  Junghoon Lee
This paper develops a standardized FHIR interaction mechanism, in which a server manages data streams created from wearable healthcare devices, aiming at prommoting the implementation of smart healthcare services. With the well-defined resource exchange protocol, the first client retrieves and displays ECG streams, which are stored as a series of FHIR observation objects, so that one process prefetch the next segment while the other is displaying the current one. Two processes cooperate through the dual buffer with the underlying synchronization method. Moreover, the second client builds a DNN-based prediction model, which detects abnormalities of the server-stored streams, achieving an accuracy of 95.4 % for the well-known PTB DB data set.
The Transactions of the Korea Information Processing Society, Vol. 14, No. 11, pp. 865-870, Nov. 2025
https://doi.org/10.3745/TKIPS.2025.14.11.865
FHIR Standard ECG Patch Device Object stream Interaction mechanism
Design and Implementation of a Medical Device Security Pre-Verification Simulator Utilizing SBOM and Security Data Integration
Woo Jung Hyun  Kwang-Man Ko
This paper proposes the design and implementation of a simulator system that enables pre-verification of medical device software security by integrating SBOM (Software Bill of Materials) with real-world vulnerability data. The proposed system generates an SBOM from user-registered software assets and automatically identifies vulnerabilities by linking with public sources such as CVE, CWE, NVD, and GitHub Advisory databases. It then derives threat scenarios based on CWE types and performs quantitative risk assessment using CVSS scores. The results are connected to a security checklist, visualized, and exported as a final report. This approach allows medical device developers to conduct structured security reviews in the early design phase and supports regulatory documentation preparation. The study demonstrates the potential of an automated analysis framework based on actual vulnerabilities to support embedded cybersecurity in medical devices.
The Transactions of the Korea Information Processing Society, Vol. 14, No. 11, pp. 871-879, Nov. 2025
https://doi.org/10.3745/TKIPS.2025.14.11.871
Digital healthcare Digital Medical Product Security Evaluation SaMD Verification Security Simulator SBOM Analysis CWE/CVE Mapping
EHD-YOLO: An Enhanced YOLOv8-Based Model for Halibut Detection in Complex Environments
Seongwang Kang  Han Kyu Lim  Hyun Seung Son
In halibut aquaculture environments, object detection is particularly challenging due to multiple factors such as fish overlapping, optical distortions caused by water surface waves, and the occurrence of bubbles. Especially for species like halibut, which have a broad body shape and a high frequency of overlapping, conventional object detection models often exhibit limitations in accuracy. To address these problems, this paper proposes EHD-YOLO (Enhanced Halibut Detection-YOLO), an improved model that integrates LSKA (Large Separable Kernel Attention), CBAM (Convolutional Block Attention Module), and CARAFE (Content-Aware ReAssembly of FEatures) modules into YOLOv8, while modifying the feature pyramid structure into an MCFP (Multi-Convolution Focused Pyramid) to enhance detection performance. The proposed model demonstrates robust detection capability not only for overlapping halibut but also against visual noise introduced by water surface waves and bubbles. For performance evaluation, feature maps generated at different pyramid levels were utilized, and comparative experiments were conducted against baseline models such as YOLOv8n, CCS-YOLOv8, and CBR-YOLO. The results show that the EHD-YOLO P234 model, which employs lower pyramid levels, reduced the number of parameters by 62%, but its computational cost increased by 1.64 times, leading to slower inference speed and decreased detection accuracy. In contrast, the EHD-YOLO P345 model required relatively higher computation, yet achieved superior results across key performance metrics (mAP, Precision, Recall, etc.) compared to existing models. Moreover, in real-world aquaculture environments with overlapping fish, water waves, and bubbles, the EHD-YOLO P345 model effectively reduced recognition errors, demonstrating its robustness and applicability
The Transactions of the Korea Information Processing Society, Vol. 14, No. 11, pp. 880-888, Nov. 2025
https://doi.org/10.3745/TKIPS.2025.14.11.880
Object Detection Overlapping/Occlusion Waves and Bubbles Model Improvement YOLOv8
A Unified Approach to Classifying Subtypes of All Cancers Using an Adaptive Learning Model with Focal Loss
Guiyuan Deng  Shiyang Wang  Kyungsook Han
Cancer subtyping is critical because treatment responses and side effects vary by subtype. We developed an adaptive neural network with focal loss (ANetFL) that classifies cancer subtypes across multiple cancer types using gene expression data. ANetFL automatically adjusts the learning rate, manages class imbalance by addressing minority and hard-to-classify subtypes, and identifies key genes characterizing cancer subtypes. In independent testing on 15 cancer types, ANetFL consistently achieved high accuracy across all metrics and outperformed existing methods. These findings suggest that ANetFL and the discovered key genes can support precise cancer subtype classification, facilitating the selection of more effective and personalized therapeutic strategies.
The Transactions of the Korea Information Processing Society, Vol. 14, No. 11, pp. 889-895, Nov. 2025
https://doi.org/10.3745/TKIPS.2025.14.11.889
Classification of cancer subtypes Neural Network Adaptive learning Focal loss
Cross-Platform Optimization of ROS 2-based Real-time Object Detection Systems
Eunjin Hwang  Seongbok Baik  Yong-Geun Hong
Real-time object detection is essential for ensuring the safety of autonomous vehicles. This study proposes three optimization techniques to improve the performance of ROS 2-based object detection services: GStreamer-based video input, client-server architecture improvement, and platform-optimized inference server implementation. On the Desktop platform, the client-server architecture improvement alone achieved over 55% performance enhancement compared to the baseline method. On the Jetson Orin Nano platform, the integrated application of all three techniques was required to achieve over 90% reduction in total processing time compared to the baseline. Furthermore, through analysis of CPU, memory, and network resources, we confirmed that the optimization techniques contribute not only to latency reduction but also to overall system efficiency improvement. This study demonstrates that optimization strategies vary depending o
The Transactions of the Korea Information Processing Society, Vol. 14, No. 11, pp. 896-906, Nov. 2025
https://doi.org/10.3745/TKIPS.2025.14.11.896
Client-Server architecture GStreamer ROS 2 Docker
Exploring User Attitudes Toward e-CNY: A Media Comment Analysis Using LDA
Yan Yan  Minjung Park  Sangmi Chai
The digital yuan (e-CNY) has become a widely discussed innovation in China's payment landscape, yet public attitudes toward it remain divided. Drawing on around 17,000 comments from Douyin, this study applied Latent Dirichlet Allocation (LDA) and SnowNLP sentiment analysis to capture user concerns and perceptions. The findings show that nearly 60% of the comments carried negative sentiment, while positive reactions were limited, indicating cautious acceptance at this stage. Three main themes emerged: first, frequent comparisons with WeChat Pay and Alipay, with many users questioning whether e-CNY is truly necessary; second, concerns over security and privacy, where government backing was seen as both a source of trust and a cause of worry about privacy; third, issues of digital inclusion, especially difficulties faced by the elderly in adapting to new payment methods. Overall, users prioritize everyday usability than about the technical sophistication of the system. These insights suggest that the future promotion of e-CNY should focus on lowering entry barriers, improving user experience, and addressing the needs of vulnerable groups to ensure broader acceptance and inclusiveness.
The Transactions of the Korea Information Processing Society, Vol. 14, No. 11, pp. 907-918, Nov. 2025
https://doi.org/10.3745/TKIPS.2025.14.11.907
Bank Digital Currency (CBDC) e-CNY sentiment analysis topic modeling User Attitudes
Lightweighting Method for DeepLabv3+ Based on MobileNet
Taejun Kim  Insu Jeong  Joowan Kang  Seungjun Jo  Byungin Moon
Recent advances in autonomous driving, medical image analysis, and surveillance technologies have led to an increasing demand for real-time image segmentation. Consequently, research on real-time segmentation methods has been actively pursued, with deep learning-based approaches demonstrating high performance metrics. However, the complex architectures, massive computational cost, and large parameter counts inherent in artificial neural networks impose significant constraints when deploying them on resource-limited embedded platforms. To address this issue, various attempts have been made, such as employing lightweight classification networks like MobileNet as backbone networks for segmentation models. Nevertheless, segmentation networks still involve substantial computational overhead and parameter size, making them difficult to apply in resource-constrained environments. Therefore, even when lightweight classification networks are used, further compression is necessary. In this paper, we propose a lightweight approach for DeepLabv3+ which MobileNetV3-Large without the last expansion layer as the backbone network. To demonstrate the effectiveness of the proposed method, we compare it with several DeepLabv3+ architectures whose backbone networks are MobileNetV3-Large, MobileNetV3-Small, MobileNetV3-Small without the last expansion layer, or MobileNetV3-Large compressed by hyperparameter tuning. Compared to the best performing architecture with MobileNetV3-Large as the backbone network, the structure with the proposed method can reduce the parameters by about 52% with only a 2.7%p performance decrease. Therefore, the proposed method effectively achieves the trade-off between performance and neural network size, which confirms that it can be practically utilized in resource-constrained environments.
The Transactions of the Korea Information Processing Society, Vol. 14, No. 11, pp. 919-924, Nov. 2025
https://doi.org/10.3745/TKIPS.2025.14.11.919
Image Segmentation DeepLabv3+ MobileNet Lightweight Resource-Constrained
A Performance Improvement of Mouse Data Protection Using CTGAN: A Method for Generation of Realistic Synthetic Mouse Data
Jinwook Kim  Kyungroul Lee
Online services delivered via computer systems require reliable user authentication for secure, non-face-to-face access. Historically, password-based user authentication has been widely used, but it remains vulnerable to credential-theft attacks such as keyloggers. Image-based user authentication methods were introduced to mitigate password exposure; however, these methods can be subverted by attacks that steal mouse input data via the WM_INPUT message, thereby neutralizing their security benefits. To counter such threats, defense techniques that inject decoy(synthetic) mouse data have been proposed. Recent advances in machine learning, however have enabled attackers to distinguish between synthetic and real mouse data with up to 99% accuracy, exposing a critical vulnerability in these defenses. To address machine learning-driven attacks on mouse data protection, prior research has applied Conditional Tabular Generative Adversarial Networks(CTGAN) to synthesize realistic decoy mouse data, demonstrating reductions in attack success rate of up to 37%. In this paper, we proposed enhancements to CTGAN-based mouse data defenses aimed at further improving protection effectiveness. Specifically, we investigate the effects of data normalization strategies, systematic hyperparameter tuning, and epoch configuration on the fidelity of generated decoy mouse data, and we introduce an optimized generation pipeline for producing high-quality synthetic mouse data. We evaluate defense performance across different data-processing and synthesis frequencies, measuring relative changes in attack success rates. Experimental results show that our proposed approach lowest attack success rates by up to 42%, representing a 5% absolute improvement over prior CTGAN-based defenses. Furthermore, our method achieves approximately and 18% improvement in average defensive performance compared to prior work. These findings demonstrate that the proposed CTGAN tuning and processing techniques materially strengthen protection of image-based user authentication data against modern machine learning-based attacks.
The Transactions of the Korea Information Processing Society, Vol. 14, No. 11, pp. 925-934, Nov. 2025
https://doi.org/10.3745/TKIPS.2025.14.11.925
Image-based Authentication Machine Learning CTGAN WM_INPUT Message
Analyzing the Applicability of AddressSanitizer for Memory Corruption Detection in TrustZone-based OP-TEE Environments
Kyungwook Boo  Byoungyoung Lee
This paper investigates the applicability of memory error detection techniques within ARM TrustZone-based Trusted Execution Environments (TEEs). While AddressSanitizer (ASAN) and Kernel AddressSanitizer (KASAN) have proven effective in general-purpose operating systems and kernels, their feasibility in restricted execution environments such as TEEs remains insufficiently explored. We experimentally evaluated ASan in the OP-TEE environment on both Trusted Applications (TAs) and the Trusted Kernel (TK), and confirmed that ASAN does not operate properly within TAs. In contrast, the stack canary mechanism was verified to function reliably in both TA and TK. To validate these findings, we performed bug injection tests targeting various memory corruption vulnerabilities, including buffer overflows, use-after-free. Furthermore, we analyzed the causes of ASAN's failure in TAs from the perspectives of execution structure, memory mapping, and loading processes. These results highlight the limitations of applying sanitizer-based runtime detection in TEEs and emphasize the need for alternative memory bug detection approaches.
The Transactions of the Korea Information Processing Society, Vol. 14, No. 11, pp. 935-939, Nov. 2025
https://doi.org/10.3745/TKIPS.2025.14.11.935
ARM Memory bug Sanitizer TrustZone TEE
A Study on an AI Evaluation Model for Improving Kiosk Accessibility for the Elderly
Choi Hyeon Seok  Kang Hyeon Seo  Ji Yu Hwan  Lim Chae Eun  Dong-Young Yoo
This study was conducted in response to the surge in demand for barrier-free kiosks following the 2025 amendment to the Anti-Discrimination against Persons with Disabilities Act. Given the limitations of the existing government certification process— characterized by prolonged duration (over two months), high costs, and the risk of evaluation failure—the research sought to explore a practical alternative solution. Grounded in the philosophy of Transgenerational Design (TD), the study aimed to develop an AI-based persona simulation system that reflects the cognitive and motor constraints of older adults, thereby establishing a framework for quantitatively evaluating kiosk usability in advance. Methodologically, the Proximal Policy Optimization(PPO) reinforcement learning algorithm was employed to train user interaction policies. Based on real usage logs collected from older adults (N=9) at welfare centers in Seoul, behavioral parameters such as slow movement speed, hand tremors, and overshoot errors were modeled and incorporated into the simulation environment. Furthermore, to replace conventional observational studies, a novel “Gaze-path Visualization Technique” was proposed, which records gaze and click events as visual traces, enhancing the intuitiveness of log-based analysis. The results of this study present a pre-assessment framework that integrates the Gaze-path Visualization Technique with PPO reinforcement learning, while also implementing an LLM-based feedback module capable of generating intuitive UI/UX improvement guidelines and automated reports. Beyond merely ensuring regulatory compliance, this system defines practical usability as a core evaluation criterion, thereby significantly reducing pre-release risks and assessment costs. Ultimately, it offers a viable solution for promoting a culture of accessibility-centered quality management.
The Transactions of the Korea Information Processing Society, Vol. 14, No. 11, pp. 940-949, Nov. 2025
https://doi.org/10.3745/TKIPS.2025.14.11.940
AI Evaluation Model Evaluation Cost Gaze-path Visualization Technique
Research on the Use of Multimodal Data for Detecting Emergency Situations Involving Elderly People Living Alone
Suyeon Lim  Seongbok Baik  Yong-Geun Hong
This study developed a multimodal anomaly detection model to monitor the safety of elderly people living alone in an ageing society, and compared and analysed its performance with that of a single modality model. The data used in the experiment was a risk detection dataset for elderly care provided by AI Hub, which included three sensor modalities: biosignals, emergency requests, and infrared images. The 1D-CNN model showed the best performance (accuracy 99.74%, F1-score 0.9975) for biosignal data, while the LSTM model recorded the highest performance (accuracy 99.99%, F1-score 0.9750) for emergency request data. The performance of the infrared image-only model (accuracy 88.01%, F1-score 0.8668) was relatively low compared to other models, but the application of the multimodal early fusion model significantly improved the anomaly detection performance based on infrared images (accuracy 94.52%, F1-score 0.9475). This experiment experimentally demonstrated that multimodal fusion effectively compensates for the limitations of a single sensor and contributes to improving the detection performance of image data.
The Transactions of the Korea Information Processing Society, Vol. 14, No. 11, pp. 950-959, Nov. 2025
https://doi.org/10.3745/TKIPS.2025.14.11.950
Multimodal Data Anomaly Detection Early Fusion
A Lightweight SMC-SORT Based Method for On-Device Real-Time Multiple Object Tracking
Lee Won Kyoung  Kim Dongho
Efficient multiple object tracking (MOT) in video is crucial for applications such as autonomous driving and video surveillance. Although recent advances in deep learning have popularized the tracking-by-detection paradigm, coupling high-accuracy detectors with trackers often hinders real-time operation on resource-constrained edge devices. This paper proposes a MOT method that attains real-time performance on edge platforms by combining a non–deep-learning Gaussian Mixture Model (GMM)-based background subtraction detector with a newly designed lightweight tracker. Building on the simplicity of SORT—i.e., a Kalman filter with Hungarian assignment—the proposed tracker introduces update and error-handling mechanisms tailored to the characteristics of background-subtracted detections. As a result, object identities are preserved through short-term occlusions and temporary stops, improving temporal continuity. Experiments demonstrate that the method achieves real-time processing with markedly lower computational load than deep learning pipelines such as YOLOv5s+DeepSORT, while delivering improved accuracy over SORT. Evaluations on Raspberry Pi 5 and Jetson AGX Orin in both CPU and GPU modes further confirm that the proposed approach offers an effective balance of efficiency and accuracy for resource-constrained edge environments.
The Transactions of the Korea Information Processing Society, Vol. 14, No. 11, pp. 960-966, Nov. 2025
https://doi.org/10.3745/TKIPS.2025.14.11.960
Multiple Object Tracking Background subtraction Gaussian mixture model SORT Edge Computing
Node Embedding Unlearning for Efficient Node Deletion in Graph Convolutional Networks
Lee Jiwon  Lee Ki Yong
Machine unlearning has emerged as a crucial concept to address data deletion requests arising from privacy protection and legal regulations, aiming to completely remove the influence of specific data from a trained model. Graph unlearning extends this concept to graph-structured data, where additional challenges arise from the need to account for relationships between nodes and edges. This study proposes a Node Embedding Unlearning method for Graph Convolutional Network (GCN)-based models that efficiently removes deleted nodes and their related information without performing full retraining. The proposed approach leverages the embeddings obtained during the initial training to recompute only the modified parts and utilizes sparse matrices containing information directly associated with the deleted nodes, thereby significantly reducing computational overhead. Experimental results demonstrate that the proposed method maintains comparable accuracy to full retraining while substantially reducing execution time. In particular, it proves effective in environments with low node deletion ratios, where real-time adaptation to graph changes is required.
The Transactions of the Korea Information Processing Society, Vol. 14, No. 11, pp. 967-974, Nov. 2025
https://doi.org/10.3745/TKIPS.2025.14.11.967
Machine Unlearning Graph Unlearning Graph Convolutional Network Graph Neural Network Graph Embedding
A Machine Learning Method for Detecting Concept Drift in Data Streams
Daewon Kim  Ji-Ho Kim  Hyo-Sang Lim
In this paper, we propose a method to detect concept drift in data stream by using machine learning techniques. Existing concept drift detecting methods that using convolutional neural networks presuppose unusual situations where labels are provided in the data stream. Also, existing methods does not consider consistency of concept. For that reason, existing methods have a problem in that input of the data stream are falsely detected as a concept drift due to sensitive response even if there is not a large difference. Therefore, we propose a technique to generate labels through machine learning in the usual data streams, and label adjust method that considers the consistency of concepts by quantifying the difference of labels in order to reduce the detection error. Our proposed method patternize inputs of data stream with autoencoder and clustering technique, and train a convolutional neural network model. Since then we detect concept drift by applying label adjust method to the output of model about past and present input.
The Transactions of the Korea Information Processing Society, Vol. 14, No. 11, pp. 975-985, Nov. 2025
https://doi.org/10.3745/TKIPS.2025.14.11.975
concept drift detection Machine Learning data stream

Forms

Search
(IN TITLE, AUTHOR, ABSTRACT,KEYWORDS)

Advanced Search ▼

POPULAR KEYWORDS
(TOP 10 KEYWORDS)

Recent Publications
(LAST 3 YEARS)

Old Journals

Indexing

All publications of TKIPS is indexed in DOI, EBSCO, Google Scholar, Crossref, and CrossCheck.

TKIPS is also selected as the Journal for Accreditation by NRF (National Research Foundation of Korea).

Related Journals