Earticle

Identity-aware activity recognition is a key enabler for customized services. However, joint modeling of activity recognition and user identification from wireless signals remains underexplored. This work presents a dual-task graph model for millimeter-wave (mmWave) frequency-modulated continuous-wave (FMCW) radar point-cloud sequences. We construct directed graphs that capture a user’s spatial structure and motion over time. A shared graph neural backbone processes these graphs and produces node embeddings that encode local spatial features and short-term dynamics. Each task-specific head first aggregates node embeddings into a graph-level representation and then performs activity or identity classification. Experiments on two public datasets demonstrate that the proposed scheme achieves classification performance comparable to single-task baselines for both activity recognition and user identification while maintaining low-latency inference. Codes are available at?https://github.com/junyongeom/mmActId/.

Stock price forecasting is a crucial challenge in FinTech industries, with implications that extend to algorithmic trading. In recent studies, Graph Neural Networks (GNNs) are employed for prediction; however, they are still limited to involving macroeconomic contexts. Hence, in this study, a novel GNN-based method for stock price forecasting is designed, with a graph building structure influenced by macroeconomic variables, namely inflation, interest rate, and GDP growth regimes. Our model captures the relationships between stocks on the basis of regime-specific, macro-driven static graphs along with an LSTM model. The proposed approach outperforms existing methods and provides a new viewpoint on stock forecasting.

As urban networks integrate heterogeneous radio access technologies (RATs), such as Wi-Fi and 5G/B5G, modeling performance becomes challenging due to interference, spatial variability, and propagation conditions. This paper proposes a graph neural network (GNN)-based framework for predicting throughput, delay, and jitter in multi-RAT environments, considering RAT type. The model encodes network topology and channel characteristics using node and edge features, capturing spatial configuration, congestion, and line-of-sight (LoS) versus non-line-of-sight (NLoS) conditions. The results show that GNNs exhibit robustness across station densities and spatial conditions. The message-passing GNN method performs well for throughput and delay, while non-graph methods better estimate jitter.

Distributed handover (HO) strategy with low complexity can provide seamless communication in low earth orbit (LEO) satellite networks. However, it is difficult to consider load balancing in distributed HO strategy, which may results in HO failures. In this paper, we propose a graph neural network (GNN) based distributed HO strategy for LEO satellite communication to maximize sum rate by considering load balancing. We first propose target satellite selection method with GNN where each user equipment (UE) selects target satellite and requests HO to it. We then employ ACK decision policy to strictly satisfy load balancing of satellites where each satellite decides HO requests from UEs depending on its load condition. To validate the proposed GNN based HO, we use the System Tool Kit (STK) for modeling LEO satellites with 22 orbits and 72 satellites are in each orbit, and evaluate the HO process during 2400 s. From this constellation, we generate 9,600 samples by randomly deploying UEs on the ground and use them as dataset. Simulation results show that the proposed GNN based HO strategy outperforms conventional HO strategies by selecting an appropriate target satellite. We also demonstrate that load balancing is satisfied due to ACK decision policy and the scalability of proposed GNN architecture is ensured with different network sizes.

In this letter, we are the first to focus on the issue of reliable and flexible radio resource allocation (RRA) for beam-hopping (BH) in satellite internet network (SIN). The main new challenges are accurate and dynamic radio resource modeling and high-efficiency RRA in high-dynamic scenarios. Therefore, we propose a novel RRA scheme for BH-SIN based on graph mapping and generative adversarial network (GAN). In our scheme, the characteristics of radio resources are first to be converted to graphical features. Then, the former RRA schemes are modeled as one of the adversarial objects, to adaptively optimize the next best solution to the changeable scenarios and situations. The simulation results show that the proposed RRA schemes improve the throughput and quality of service by 15% and 22%.

When developing a classifier using various objective functions, it is important to compare the performances of the classifiers. Although there are statistical analyses of objective functions for classifiers, simulation results can provide us with direct comparison results and in this case, a comparison criterion is considerably critical. A Receiver Operating Characteristics (ROC) graph is a simulation technique for comparing classifiers and selecting a better one based on a performance. In this paper, we adopt the ROC graph to compare classifiers trained by mean-squared error, cross-entropy error, classification figure of merit, and the n-th order extension of cross-entropy error functions. After the training of feed-forward neural networks using the CEDAR database, the ROC graphs are plotted to help us identify which objective function is better.

This paper addresses the critical task of anomaly detection in river network sensor data, essential for accurate and continuous water quality monitoring. We propose M-MAD (Multi-Modal Anomaly Detection), a novel approach that integrates multi-modal features, including sensor data, weather information, and historical anomalies. M-MAD builds on the Graph Deviation Network (GDN) framework by introducing an improved anomaly threshold criterion derived from the learned graph structure. Our evaluation employs rigorous benchmarking simulations that mimic complex dependency structures and diverse anomalies, thoroughly assessing the strengths and weaknesses of M-MAD compared to existing methods. Results demonstrate M-MAD's superior performance in handling high-dimensional datasets and its enhanced interpretability, crucial for effective anomaly detection.

Predicting accurate human migration patterns is crucial for effective urban planning. However, accurate human migration patterns prediction remains a challenging task. Existing methods, such as Graph Neural Network approaches, often overlook dynamic temporal variations and directional dependencies in large-scale migration data. To overcome this challenge, we propose MiGA-Net (Migration Graph Attention Network), a graph Neural network-based framework enhanced with an attention mechanism to capture complex spatiotemporal dependencies and highlight significant migration flows on the domestic and international level. We utilize two different datasets of the Shinan-gun, South Korea, for international and domestic regions. Experimental results show that the proposed MiGA-Net achieved superior performance over both datasets. The model achieved 0.0027 MAE for domestic flow and 0.0155 for international flow, demonstrating the effectiveness of the proposed framework.

코로나19로 인해 생활양식이 오프라인에서 온라인으로 전환되는 추세를 보이고 있다. 그 추세중 하나로써 온라인 상에서 사회적 교류를 나눌 수 있는 라이브 스트리밍 플랫폼의 수요도 증가하고 있고 수익 또한 상승했다. 라이브 스트리밍 플랫폼의 주 수익은 재능있는 스트리머들과의 제휴 관계로부터 창출되기에 플랫폼 입장에서는 향후 제휴 관계를 형성하게 될 성장 가능성이 높은 스트리머를 사전에 발굴하여, 이들을 지속적으로 관리하는 것이 중요하다. 이에 본 연구에서는 플랫폼과 스트리머간의 제휴 여부에 대해 예측하고자 소셜네트워크 데이터를 활용한 그래프 신경망(Graph Neural Network) 방법론의 적용을 제안해 보고자 한다.

이 연구는 네트워크 의제설정이론을 기반으로 유튜브 내 인공지능 담론의 구조적 특성과 변화 양상을 파악한다. 2018년부터 2025년 3분기까지 게시된 관련 영상 29,697건을 수집하고, 그래프 신경망 기반 계층적 이질 그래프 트랜스포머 모델을 활용하여 의미적 유사성, 구조적 공출현, 생산자 맥락의 상대적 영향력을 분석했다. 분석 결과, 생산자 맥락인 채널이 의제 형성에 가장 큰 영향을 미쳤으며, YTN 등 레거시 미디어가 의제 네트워크 허브로 기능했다. 담론 구조는 거시 의제에서 미시 생활 의제로 확산하는 계층적 연결성을 보였고, 시계열적으로는 전망, 혁신, 일상화로 누적 진화하는 안정적 생태계를 확인했다. 이 연구는 디지털 플랫폼 내 의제설정 구조를 규명하고, 레거시 미디어의 게이트키핑 기능이 디지털 플랫폼 환경에 서도 유효함을 실증했다. 더불어, 그래프 신경망 접근을 담론 분석에 적용하여 융합 연구 방법론의 효용성을 입증했다는 데 의의가 있다.

This study applies Network Agenda-Setting theory to examine the structural characteristics and temporal evolution of artificial intelligence discourse on YouTube. A total of 29,697 related videos published between 2018 and the third quarter of 2025 were collected, and a graph neural network-based hierarchical heterogeneous graph transformer model was employed to analyze the relative influence of semantic similarity, structural co-occurrence, and producer context. The results indicate that producer context, represented by YouTube channels, exerts the strongest influence on agenda formation, with legacy media such as YTN functioning as hubs within the agenda network. The discourse structure exhibits hierarchical connectivity through which macro-level agendas diffuse into micro-level everyday agendas, and a temporally stable ecosystem was confirmed, characterized by cumulative evolution across three phases of anticipation, innovation, and normalization. This study contributes by identifying the agenda-setting structure within a digital platform environment and empirically demonstrating that the gatekeeping function of legacy media remains effective in digital platform-based contexts. It additionally advances convergent research methodology by illustrating the analytical utility of graph neural network approaches in discourse analysis.

의료기기 소프트웨어에서 발생한 취약점은 환자의 안전과 직결되며, 최근 소프트웨어 공급망의 복잡성이 증가함에 따라 Software Bill of Materials(SBOM) 기반 취약점 관리의 중요성이 더욱 강조되고 있다. 특히, U.S. Food and Drug Administration(FDA)은 의료기기 인허가 과정에서의 SBOM 제출 및 체계적인 취약점 관리 체계 구축을 요구하고 있으나, 기존의 취약점 관리 방법은 주로 Common Vulnerability Scoring System(CVSS) 기반 정적 위험 평가 혹은 Software Composition Analysis(SCA) 중심의 취약점 식별에 의존하고 있어, 소프트웨어 구성 요소 간 의존성 관계 및 실제 악용 가능성이 높은 취약점 정보를 충분히 반영하지 못하는 한계가 존재한다. 따라서 본 연구에서는 SBOM 데이터를 기반으로 취약점 간 의존성을 반영하여 이종 그래프(Heterogeneous Graph)를 구축하고, 그래프 신경망(Graph Neural Network, GNN)을 활용하여 컴포넌트 단위의 취약점 위험도를 정량적으로 평가하는 방법을 제안한다. 제안 기법은 CVSS 점수, Known Exploited Vulnerabilities(KEV) Catalog, 그리고 컴포넌트 간 의존성 구조를 통합적으로 반영하여 위험도를 산출하고, 이를 기반으로 패치 우선순위를 자동으로 결정한다. 실험 결과, 제안 모델은 기존의 정적 평가 기반 접근 방식 대비 주요 ranking 성능 지표에서 전반적으로 우수한 성능을 보였으며, 의료기기 소프트웨어의 실질적인 취약점 대응 우선순위 결정에 효과적으로 활용될 수 있음을 확인했다.

Vulnerabilities in medical device software are directly linked to patient safety, and the increasing complexity of software supply chains has amplified the importance of Software Bill of Materials (SBOM)-based vulnerability management. In particular, the U.S. Food and Drug Administration requires SBOM submission and systematic vulnerability management as part of the medical device approval process. However, existing approaches primarily rely on static risk assessment based on the Common Vulnerability Scoring System or vulnerability identification using Software Composition Analysis(SCA), which fail to sufficiently capture dependency relationships among software components and real-world exploitability. To address this limitation, this paper proposes a method for component-level vulnerability risk assessment by constructing a heterogeneous graph from SBOM data and applying Graph Neural Networks(GNNs). The proposed approach integrates CVSS scores, the Known Exploited Vulnerabilities Catalog, and dependency structures to compute risk scores and automatically prioritize patches. Experimental results show that the proposed model outperforms conventional static approaches across key ranking metrics, demonstrating its effectiveness for practical vulnerability prioritization in medical device software.

차세대컴퓨팅의 발달로 인하여 인공지능은 다양한 분야에 적용되어 놀라운 성능을 보인다. 하드웨어 트로이목마는 하드웨어의 설계, 제조 과정에서 주입될 수 있는 악성 변조 행위로 보안 키 유출, 시스 템 오류 등의 원인이 된다. 이러한 트로이목마 위협에 대응하여 기존의 연구들은 하드웨어 디자인 내 의 여러 요소에서 그래프 생성 및 그래프 신경망을 구축하여 인공지능을 통한 하드웨어 트로이목마 검출 방법을 고안하였다. 본 논문에서는 기존 연구들의 하드웨어 트로이목마 검출 방법을 비교한다.

고정밀 지도의 갱신은 정사영상 또는 점군 데이터 등을 원천 자료로 하여 기하 정보를 우선적 으로 수정한 이후 지도를 구성하는 공간객체들 간의 연관관계를 재정립하는 방식으로 진행된다. 이러한 일련의 과정들은 기하 정보를 처리하는 데에 많은 시간을 소요하므로 차량의 실시간 경 로 계획(Real-time route planning)에 빠르게 적용되기 어렵다. 따라서 이 연구에서는 그래프 구조 를 활용하여 경로 계획을 위한 도로 연결구조를 우선적으로 업데이트 하는 방식 및 도로 네트워 크의 특징을 고려한 그래프 구조의 저장 유형을 제안하였다. 또한 제안된 방법을 실제 도로 자료 에 적용해 봄으로써 실시간 경로 정보 전송 시의 활용 가능성에 대해 검토하였다.

The update of a high-precision map was carried out by modifying the geometric information using ortho-images or point-cloud data as the source data and then reconstructing the relationship between the spatial objects. These series of processes take considerable time to process the geometric information, making it difficult to apply real-time route planning to a vehicle quickly. Therefore, this study proposed a method to update the road network for route planning using a graph data structure and storage type of graph data structure considering the characteristics of the road network. The proposed method was also reviewed to assess the feasibility of real-time route information transmission by applying it to actual road data.

본 연구에서는 그래프 신경망 모형을 포함한 6개의 기계학습모형을 활용하여 한국 기업의 신용변동을 예측했다. 이를 예측하기 위해 수익성, 성장성, 안정성, 활동성 지표들을 활용하였고, 주성분 분석을 진행하여 지표별로 두 개씩 활용했다. 또한, 신용등급변동이 없는 경우가 대다수인 불균형 데이터를 처리하기 위해 SMOTE 방법 론을 활용하여 데이터를 오버샘플링했다. 본 연구에서는 산업구조를 반영하기 위해 이를 나타내는 상장시장, 데이터 시점, 현재 신용등급과 업종의 네 분류로 나누어 모형을 만들고 이를 최종적으로 결합하여 예측 결과를 도출했다. 결과를 살펴보면, 모든 모형이 데이터 비중에 맞게 선택하는 벤치마크 모형보다 우수하고, 그래프 신경 망 모형이 특히 더 우수한 성능을 나타내는 것을 확인할 수 있다. 또한, 변수 중요도를 분석한 결과, 성장성 지표와 활동성 지표가 중요한 것을 확인할 수 있었고, 그래프 신경망에서는 업종 그래프와 상장시장 그래프가 성능이 좋은 것을 확인할 수 있었다.

Credit rating is an essential component of the financial sector, as it evaluates a firm’s capacity to meet debt obligations. In the Republic of Korea, rating agencies assign levels ranging from AAA to D, with additional modifiers, and these ratings significantly affect financing conditions. Traditional methods typically rely on logistic regression and selected financial variables, yet these approaches often face difficulties in capturing the intricate or nonlinear patterns present in corporate financial data. In response to these challenges, researchers have increasingly turned to advanced machine learning algorithms that can account for more complex relationships. Nevertheless, their deployment is limited by relatively small datasets—particularly among smaller firms—and by concerns regarding model interpretability. The present study proposes a machine learning framework, including a Graph Neural Network (GNN), to predict rating changes in Korean firms. The data set spans 2010 to mid-2024 and includes 182 firms, yielding 1,417 year-level samples. Each annual observation is labeled according to whether its credit rating was upgraded (+1), downgraded (–1), or left unchanged (0). Because most observations lie in the unchanged category, the data are highly imbalanced. To address this imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) is employed, generating additional samples in the minority classes. In addition, Principal Component Analysis (PCA) is utilized to reduce the dimensionality of sixteen indicators representing profitability, growth, stability, and activity. Six models are assessed: logistic regression, LASSO, random forest, support vector machine, Light Gradient Boosting Machine (LGBM), and a GNN. The GNN-based approach is noteworthy for modeling financial indicators or corporate attributes as nodes within a graph, with edges delineating the relationships among these nodes. Such a representation allows for the capture of latent dependencies that are difficult to detect in methods that treat predictors independently. Furthermore, each corporate sample belongs to discrete subgroups determined by listing market (KOSPI or KOSDAQ), data period, rating band, and industry classification. The proposed hierarchical GNN merges these subgroup-specific networks to produce a consolidated prediction, thereby incorporating both firm-level attributes and group-level characteristics. When compared to a benchmark that classifies samples randomly in proportion to the observed class distribution, all six machine learning algorithms demonstrate superior performance. The GNN shows the highest precision and F1-scores, suggesting that it is particularly effective at identifying upgrades and downgrades, which are far less common than no rating changes. Nonetheless, like other models, it finds rare rating shifts more challenging to predict, highlighting the impact of data imbalance and the difficulty of forecasting uncommon events. An inspection of feature importance across models underscores the significance of growth and activity metrics, implying that sales expansion, equity growth, and the efficient use of assets offer robust signals of rating volatility. Moreover, the GNN indicates that distinguishing firms by industry group is especially influential, possibly because each sector’s distinctive regulatory, economic, and financial traits shape its credit risk profile. Compared to certain deep neural networks that demand extensive datasets, the GNN-based method presented here is relatively more practical in settings with limited data, including smaller firms with incomplete rating histories. Additionally, this approach provides improved transparency, as the graph architecture clarifies how different financial indicators or subgroups collectively affect rating transitions. Future work may benefit from enlarging the dataset, experimenting with alternative oversampling strategies such as ADASYN, and examining cost-sensitive learning to mitigate the imbalance problem further. Investigations might also consider alternative graph structures that connect entire firms as nodes and delineate inter-firm relationships or incorporate advanced architectures such as Transformers or LSTM networks. In summary, the findings suggest that a GNN-based framework can improve credit rating predictions by capturing complex interactions that traditional or other advanced machine learning methods may overlook. While data imbalance is still a problem, its consequences are somewhat mitigated by SMOTE. The significance of growth, activity, and sector-specific characteristics suggests that more accurate and comprehensible rating projections can be produced by integrating richer and more interconnected data. In the end, more investigation and more extensive data gathering should improve the precision and dependability of credit rating systems, leading to a better comprehension of the dynamics of corporate finance.

교통 혼잡을 해결하기 위한 AI 기반 속도 예측 연구는 활발하게 진행되고 있다. 하지만, 인공지능의 추론 과정을 설명하는 설명 가능한 AI의 중요성이 대두되고 있는 가운데 AI 기반 속도 예측의 결과를 해석하고 원인을 추리하는 연구는 미흡하였다. 따라서 본 논문에서는 `설명 가능 그래프 심층 인공신경망 (GNN)`을 고안하여 속도 예측뿐만 아니라, GNN 모델 입력값의 마스킹 기법에 기반하여 인근 도로 영향력을 정량적으로 분석함으로써 혼잡 등의 상황에 대한 추론 근거를 도출하였다. TOPIS 통행 속도 데이터를 활용하여 서울 시내 혼잡 도로를 기준으로 예측 및 분석 방법론을 적용한 후 영향력 높은 인근 도로의 속도를 가상으로 조절하는 시뮬레이션 통하여 혼잡 도로의 통행 속도가 개선됨을 확인하여 제안한 방법론의 타당성을 입증하였다. 이는 교통 네트워크에 제안한 방법론을 적용하고, 그 추론 결과에 기반한 특정 인근 도로를 제어하여 교통 흐름을 개선할 수 있다는 점에 의미가 있다.

AI-based speed prediction studies have been conducted quite actively. However, while the importance of explainable AI is emerging, the study of interpreting and reasoning the AI-based speed predictions has not been carried out much. Therefore, in this paper, 'Explainable Deep Graph Neural Network (GNN)' is devised to analyze the speed prediction and assess the nearby road influence for reasoning the critical contributions to a given road situation. The model's output was explained by comparing the differences in output before and after masking the input values of the GNN model. Using TOPIS traffic speed data, we applied our GNN models for the major congested roads in Seoul. We verified our approach through a traffic flow simulation by adjusting the most influential nearby roads' speed and observing the congestion's relief on the road of interest accordingly. This is meaningful in that our approach can be applied to the transportation network and traffic flow can be improved by controlling specific nearby roads based on the inference results.

기존의 네트워크 구조는 게이트웨이 보안을 경계로 사용하므로 내부자 간의 무단 액세스를 방지하고 내부자 공격을 완화하기가 어렵다. 또한 조직의 규모가 커지면서 네트워크 트래픽이 복잡해지면서 모니터링 및 위협 식 별이 어려운 작업이 되었다. 이러한 문제를 극복하기 위해 그래프 데이터베이스를 사용하여 네트워크 노드를 모 델링하고 네트워크 동작을 분석할 것을 제안한다. 네트워크 트래픽 데이터를 수집하고 전처리를 통한 데이터 확 보와 그래프 모델링을 통한 내부 네트워크의 구조, 운영 및 잠재적인 보안 위협에 대한 통찰력을 얻을 수 있다. 이를 통해 접근통제 테이블과 접근통제 정책을 수립함으로써 보안 위협을 가하는 잠재적인 노드를 식별하고 비정 상적인 활동을 감지할 수 있다. 또한 접속 패턴을 분류하고 정상 편차를 식별하며 악의적인 사용을 방지하기 위 한 노드 간 접근통제 조치를 구현함으로써 발생할 수 있는 잠재적 위협을 방지할 수 있다.

Existing network structures use gateway security as a boundary, making it difficult to prevent unauthorized access between insiders and mitigate insider attacks. Additionally, as organizations grow in size, network traffic becomes more complex, making monitoring and threat identification a difficult task. To overcome these problems, we propose to model network nodes and analyze network behavior using a graph database. You can collect network traffic data, obtain data through preprocessing, and gain insight into the structure, operation, and potential security threats of internal networks through graph modeling. Through this, it is possible to identify potential nodes that pose security threats and detect abnormal activities by establishing access control tables and access control policies. In addition, potential threats that may occur can be prevented by classifying access patterns, identifying normal deviations, and implementing access control measures between nodes to prevent malicious use.