Latest Issue
Vol. 52,
No. 3,
Mar.
2025
All Issues
Lightweight Vertical Autoscaling Method Using Taylor Series for Serverless Computing
http://doi.org/10.5626/JOK.2025.52.3.181
Serverless computing has become essential in modern IT infrastructure by utilizing autoscaling to reduce server management burdens, enabling developers to concentrate on service. However, as serverless environments now handle multiple requests per instance, the limitations of horizontal autoscaling have become more apparent. This underscores the increasing need for vertical autoscaling, which dynamically adjusts the resource allocations for each instance. Traditional vertical autoscaling methods, designed for long-running cloud applications, are not well-suited for serverless environments that require rapid response and short execution times. This paper introduces a lightweight vertical autoscaling method that employs Taylor series to enhance both resource efficiency and performance. Experiments with FunctionBench demonstrate that the proposed method reduces resource reservations and wasted resource slack compared to Vertical Pod Autoscaler (VPA) and Tiny Autoscaler, while also improving average and 99th-percentile tail latency. Specifically, when compared to VPA, resource reservations and slack decreased by 18.6% and 45%, respectively, while average and tail latency improved by 31.5% and 53.8%. Additionally, it exhibited the lowest overhead, confirming its effectiveness as a lightweight autoscaling solution.
CraftGround: A Flexible Reinforcement Learning Environment Based on the Latest Minecraft
Hyeonseo Yang, Minsu Lee, Byoung-Tak Zhang
http://doi.org/10.5626/JOK.2025.52.3.189
This paper presents CraftGround, an innovative reinforcement learning environment based on the latest version of Minecraft (1.21). CraftGround provides flexible experimental setups and supports reinforcement learning in complex 3D environments, offering a variety of observational data, including visual information, audio cues, biome-specific contexts, and in-game statistics. Our experiments evaluated several agents, such as VPT (Video PreTraining), PPO, RecurrentPPO, and DQN, across various tasks, including tree chopping, evading hostile monsters, and fishing. The results indicated that VPT performed exceptionally well due to its pretraining, achieving higher performance and efficiency in structured tasks. In contrast, online learning algorithms like PPO and RecurrentPPO demonstrated a greater ability to adapt to environmental changes, showing improvement over time. These findings highlight CraftGround's potential to advance research on adaptive agent behaviors in dynamic 3D simulations.
Improved Software Defect Prediction with Gated Tab Transformer
Saranya Manikandan, Duksan Ryu
http://doi.org/10.5626/JOK.2025.52.3.196
Software Defect Prediction (SDP) plays a crucial role in ensuring software quality and reliability. Although, traditional machine learning and deep learning models are widely used for SDP, recent advancements in the field of natural language processing have paved the way for applying transformer-based models in software engineering tasks. This paper investigated transformer-based model as a potential approach to improve SDP model quality, ultimately aiming to enhance software quality and optimize testing resource allocation. Inspired by the Gated Tab Transformer’s (GTT) ability to effectively model relationship within features, we evaluated its effectiveness in SDP. We conducted experiments using 15 software defect datasets and compared results with other state-of-the-art machine learning and deep learning models. Our experiments showed that GTT outperformed state-of-the-art machine learning models in terms of recall, balance, and AUC (increase by 42.1%, 10.93%, and 7.1%, respectively). Cohen's d confirmed this advantage with large and medium effect sizes for GTT on these metrics. Additionally, an ablation study assessed the impact of hyperparameter variations on performance. Thus, GTT's effectiveness address the challenges of SDP, potentially leading to more effective testing resource allocation and improved software quality.
IDFusion: Joint Angle Measurement Method through Fusion of Inertial Measurement Sensor and Depth Camera
Juyeon Park, Mingyu Park, Gyumin Park, Hyun Lee
http://doi.org/10.5626/JOK.2025.52.3.208
Recent advancements in human and object recognition technologies are increasingly applied across various fields, particularly in motion detection research utilizing inertial measurement sensors and depth cameras in areas such as gaming, healthcare, and security. However, challenges such as cumulative errors and variable measurement accuracies depending on the environment persist. This study proposed IDFusion, a method that could integrate inertial measurement sensors and depth cameras for joint angle measurement, distinguishing itself through data transformation and joint angle conversion stages before fusion. Comparative analysis against using inertial measurement sensors and depth cameras individually demonstrated a superior performance of IDFusion. This technique holds promise for applications in healthcare, sports science, and human-computer interaction.
Geographical Adaptive Attention Model for Points of Interest Recommendation
Muyeon Jo, Sejin Chun, Jungkyu Han
http://doi.org/10.5626/JOK.2025.52.3.217
Geographical influence, stemming from the location of Points of Interest (POIs), plays a vital role in POI recommendation. Most current studies utilize geographical information such as distance and location to define and extract POI-specific geographical influences for personalized recommendations. These approaches primarily emphasize distance-based influence, which gauges user preferences based on proximity, while often overlooking area-based influence, which reflects preferences for regions with specific POI characteristics. This paper introduces a POI recommendation model based on an attention network that integrates both distance- and area-based influences. The model adaptively assesses how previously visited POIs impact the likelihood of visiting a target POI, taking into account regional characteristics and user preferences. Experiments conducted on real-world datasets indicate that the proposed method significantly outperforms baseline models, achieving improvements of approximately 6–12% in Prec@10, 8–10% in Recall@10, and 6–7% in HR@10.
Improvement of XRCE-DDS Communication System for Swarm Environment of Unmanned Vehicles Based on PX4-ROS2
Hyeongyu Lee, Doyoon Kim, Dongoo Lee, Sungtae Moon
http://doi.org/10.5626/JOK.2025.52.3.227
Recently, swarm vehicles are being used in various fields due to the development of swarm operation technology. Among various systems that constitute a swarm vehicle, PX4-ROS2 connects the PX4, an unmanned vehicle control computer, and ROS2 for mission execution through XRCE-DDS (eXtremely Resource Constrained Environments-Data Distribution Service), an open-source-based software that supports real-time communication between devices. However, the operation of swarm unmanned vehicles based on a wireless network using a distributed service of XRCE-DDS is not optimized. It requires communication optimization work for stable operation. In this paper, we analyzed the XRCE-DDS communication structure operating in PX4-ROS2 and proposed a new Discovery mechanism to solve the problem of increased communication volume due to increased nodes during swarm operation. We present a method to enhance the stability and scalability of communication and verified it through simulation.
Drug Toxicity Prediction Using Integrated Graph Neural Networks and Attention-Based Random Walk Algorithm
Jong-Hoon Park, Jae-Woo Chu, Young-Rae Cho
http://doi.org/10.5626/JOK.2025.52.3.234
The traditional drug development process is often burdened by high costs and lengthy timelines, leading to increasing interest in AI-based drug development. In particular, the importance of AI models for preemptively evaluating drug toxicity is being emphasized. In this study, we propose a novel drug toxicity prediction model, named Integrated GNNs and Attention Randon Walk (IG-ARW). The proposed method integrates various Graph Neural Network (GNN) models and uses attention mechanisms to compute random walk transition probabilities, extracting graph features precisely. The model then conducts random walks to extract node features and graph features, ultimately predicting drug toxicity. IG-ARW was evaluated on three different datasets, demonstrating strong performances with AUC scores of 0.8315, 0.8894, and 0.7476, respectively. Notably, the model was proven to be highly effective not only in toxicity prediction, but also in predicting other drug characteristics.
A Diffusion-based Trajectory Prediction Model for Flight Vehicles Considering Pull-up Maneuvers
Seonggyun Lee, Joonseong Kang, Jeyoon Yeom, Dongwg Hong, Youngmin Kim, Kyungwoo Song
http://doi.org/10.5626/JOK.2025.52.3.241
This paper proposes a new model for processing multivariate time series data aimed at predicting nonlinear trajectories related to aircraft pull-up maneuvers. To achieve this, aircraft trajectories were predicted using CSDI (Conditional Score-based Diffusion Models for Imputation), a state-of-the-art generative AI model. Specifically, because the flight distance and shape of the aircraft vary significantly depending on the presence of pull-up maneuvers, the data were separated into subsets with and without these maneuvers to train and predict distinct models. Experimental results demonstrated that the model predicted trajectories very similar to actual trajectories and achieved superior performance in MAE, RMSE, and CRPS metrics compared to existing deep learning models. This study not only enhances the accuracy of aircraft trajectory prediction but also suggests the potential for more sophisticated predictions through future integration with Classifier Diffusion models.
SyllaBERT: A Syllable-Based Efficient Robust Transformer Model for Real-World Noise and Typographical Errors
Seongwan Park, Yumin Heo, Youngjoong Ko
http://doi.org/10.5626/JOK.2025.52.3.250
Training a Korean language model necessitates the development of a tokenizer specifically designed for the unique features of the Korean language, making this a crucial step in the modeling process. Most current language models utilize morpheme-based or subword-based tokenization. While these approaches work well with clean Korean text data, they are prone to out-of-vocabulary (OOV) issues due to abbreviations and neologisms frequently encountered in real-world Korean data. Moreover, actual Korean text often contains various typos and non-standard expressions, to which traditional morpheme-based or subword-based tokenizers are not sufficiently robust. To tackle these challenges, this paper introduces the SyllaBERT model, which employs syllable-level tokenization to effectively address the specific characteristics of Korean, even in noisy and non-standard contexts, with minimal resources. A compact syllable-level vocabulary was created, and a syllable-based language model was developed by reducing the embedding and hidden layer sizes of existing models. Experimental results show that, despite having approximately four times fewer parameters than subword-based models, the SyllaBERT model outperforms them in natural language understanding tasks on real-world conversational Korean data that includes noise.
Efficiently Lightweight Korean Language Model with Post-layer Pruning and Multi-stage Fine-tuning
http://doi.org/10.5626/JOK.2025.52.3.260
The increasing size of large-scale language models has led to the need for lightweighting for practical applications. This study presents a method to reduce the existing 8B model to 5B by late-layer pruning, while maintaining and improving its performance through two phases of fine-tuning. In the broad fine-tuning phase, we expanded the model's ability to understand and generate Korean by utilizing English-Korean parallel data and a large Korean corpus, and in the refined fine-tuning phase, we enhanced its expressive and inferential capabilities with high-quality datasets. In addition, we integrated the strengths of individual models through model merging techniques. In the LogicKor leaderboard evaluation, the proposed model performed well in the areas of reasoning, writing, and comprehension, with an overall score of 4.36, outperforming the original Llama-3.1-8B-Instruct model (4.35). This demonstrates a 37.5% reduction in model size while still improving performance.
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