Taesu Cheong

Obstructive sleep apnea (OSA) is common yet frequently underdiagnosed, partly because overnight polysomnography (PSG) is logistically burdensome and access to specialized testing is limited. We aimed to develop machine-learning models for OSA risk screening using multimodal digital phenotyping from consumer-grade wearable devices, smartphone-based assessments, and clinical scales. We enrolled 338 participants and collected data over four weeks. After preprocessing, 107 features were derived from wearable-derived physiological and activity measures, smartphone-based records, and questionnaire-based clinical risk profiles, and used to classify high- versus low-risk OSA groups defined by the Berlin Questionnaire. Across multiple model configurations, predictive performance was high, with the best-performing model achieving an AUC of up to 0.94 and an F1 score of 0.80 in the internal validation set. Consistently influential predictors included body mass index, Insomnia Severity Index score, Smartphone Overuse Screening Questionnaire score, resting heart rate, and heart rate recovery. These findings suggest that multimodal digital phenotyping from accessible consumer technologies may support scalable pre-screening for OSA risk in real-world settings. Further validation against PSG-confirmed OSA outcomes is needed.Trial Registration: Clinical Research Information Service (CRIS) KCT0009175 (Registration data: Feb-15-2024) (https://cris.nih.go.kr/cris/search/detailSearch.do?search_lang=E&focus=reset_12&search_page=M&pageSize=10&page=undefined&seq=26133&status=5&seq_group=26133).

Background and aims: This study aimed to identify the differences in characteristics between high- and low-risk smartphone users among individuals with insomnia symptoms using digital phenotyping and clinical assessments. Methods: A total of 246 participants with insomnia symptoms ( M = 31.14, SD = 10.09) were monitored for four weeks using the smartphone application and wearable devices. The participants were divided into high-( n = 141) and low-risk ( n = 105) smartphone overuse groups based on a Smartphone Overuse Screening Questionnaire. Clinical scale results and wearable data were analyzed using ANCOVA and logistic regression, controlling for age, sex, and BMI. Results: After covariate adjustment, the high-risk group showed significantly greater biological rhythm disruption (K-BRIAN: LS-mean difference = 6.86, p < 0.000), more severe insomnia (ISI index: aOR: 2.63, p = 0.0005), and poorer sleep quality (PSQI-K: aOR: 2.41, p = 0.0015). Psychological distress, including depression (PHQ-9 index: aOR: 2.77, p = 0.0001) and anxiety (GAD-7 index: aOR: 1.59, p = 0.0059), was more pronounced in the high-risk group. Bedtime procrastination (BPS index: aOR: 1.96, p = 0.0173) and stress reactivity to insomnia (FIRST index: aOR: 1.67, p = 0.0574) were significantly elevated. Digital phenotyping revealed persistent differences in minimum daytime heart rate and exercise intensity patterns, while many activity-related measures lost significance after adjustment. Discussion and conclusions: Smartphone overuse is independently associated with severe circadian disruption, insomnia, and psychological distress. The integrated assessment approach revealed critical biomarkers and behavioral patterns. Targeted interventions focused on circadian stabilization and behavioral sleep patterns may improve sleep quality and mental health outcomes in this population. Longitudinal research is needed to establish causality.

In response to the rapid growth of e-commerce and the resulting operational challenges in urban last mile logistics, this study introduces the Multi-period Multi-depot Vehicle Routing Problem with Mobile Center (MPMDVRP-MC). The proposed model jointly determines, over multiple periods, which candidate depot sites to activate for deploying mobile centers, how to assign customers to the activated sites, and how to construct capacity-feasible delivery routes, while allowing mobile centers to be opened, closed, and repositioned across days to respond to time-varying demand.We develop a mixed-integer linear programming (MILP) formulation and a genetic algorithm-based heuristic tailored to this multi-period setting. Computational experiments on adapted Solomon benchmark instances [1] show that the heuristic matches or closely approximates MILP solutions on small instances and produces high-quality solutions on larger instances where the MILP frequently reaches the time limit or fails to prove optimality. Sensitivity analyses further reveal systematic trade-offs among travel, opening/operating, and repositioning costs, clarifying when the system behaves as a mobile, adaptive network versus a more static depot configuration. A case study using public parking-lot data from Seoul provides a proof-of-concept illustration of how repositioning can reduce travel costs under shifting demand patterns. Beyond cost efficiency, the mobile-center paradigm combined with light electric delivery modes has potential implications for reducing reliance on heavy vehicles in dense urban areas with fixed depot positions, which we discuss as a qualitative motivation.

In this paper, we propose a three-dimensional shortest path problem for exhaust pipe routing in the semiconductor fabrication industry, involving a single origin and multiple candidate destinations. While previous studies have addressed variations of the shortest path problem, this study extends it into a three-dimensional space in semiconductor fabrication factory, incorporating calculations for pressure loss due to pipe length and multiple types of bend angles, obstacle avoidance, and relevant safety standards for industrial applications. Given the increasing complexity and frequent layout changes in modern semiconductor fabrication facilities, efficient and automated pipe routing has become a crucial task to reduce rework, installation time, and safety risks. To solve this problem, we developed a mixed-integer linear programming (MILP) model. Given its computational complexity, we propose a Pressure-based A* algorithm with a Divide-and-Conquer strategy (PA*-DC), which iteratively applies the 3D Pipe Routing A* (3DP_A*) algorithm. We validated the algorithms through computational experiments using a real-world dataset provided by Samsung Electronics and compared the results with current industry de facto standards for pipe design in actual fabrication facilities. Experimental results confirm that the proposed PA*-DC algorithm achieves high-quality routes and improves pressure efficiency by up to 53.20%, demonstrating its practicality for real-world applications.

The growing interest in electric vehicles, driven by the climate crisis, has underscored the potential of battery-swapping electric two-wheeled vehicles (BSE2WVs) in overcoming the limitations of conventional rechargeable systems, expanding the global battery-as-a-service (BaaS) market. Accurate state-of-charge (SoC) prediction is essential for BaaS providers to efficiently manage battery swapping. However, most existing studies focus on four-wheeled vehicles, overlooking the unique driving patterns and dynamics of two-wheelers. Moreover, despite the adoption of machine-learning-based approaches, the cold start problem remains a challenge due to data scarcity. This study analyzes the driving patterns of BSE2WVs and proposes an SoC prediction framework tailored to these patterns. Ensemble clustering techniques identify driving patterns while considering risks and illegal behaviors, incorporating lateral movement characteristics. Training models on driver-specific patterns enhance prediction accuracy, while a data interpolation method extracts representative SoC consumption trends to mitigate data scarcity. Additionally, a skip connection method utilizing initial window data captures battery consumption variations based on the current SoC. Experiments using real-world BSE2WV data from a BaaS provider in Seoul, South Korea, demonstrate that the proposed framework outperforms existing models, achieving high prediction accuracy even under limited data conditions, with an average improvement of 29.25% through reduced MSE.

Supply chain mapping is crucial for global companies to identify and mitigate potential risks. Although natural language processing techniques are analyzed to extract supply chain maps from textual sources to automate the process, they require extensive manual annotation of training data, limiting the scalability and efficiency of these approaches. This study explores novel methods using large language models to fully automate supply chain mapping, focusing on synthetic data generation and improving relation classification techniques. This study investigates the use of large language models for relation classification in supply chain contexts and explores their potential for generating synthetic datasets. The performance of these synthetic datasets is compared with those of manually annotated datasets. The experimental results demonstrate that synthetic data outperform manual data if manual data are insufficient. To demonstrate its practical applicability, this study applies the developed method to generate a supply chain map from mining-related news articles. The subsequent supply chain map visualization for mining-related industries indicates the effectiveness of the method in automated, real-time mapping using publicly available information, enabling the rapid identification of supply network changes. This study contributes to the supply chain management and natural language processing fields by advancing sophisticated, AI-driven supply chain analysis tools.

Fires frequently occur worldwide, devastating neighborhoods and forests owing to challenges in early extinguishment. Computer vision-based deep learning models have recently been employed in fire detection research. The primary objective of this study is to develop a multiclass fire detection model using images and videos. In this study, an ensemble fire detection model, FireDetectNet (FDN), comprising several classification models (ResNeXt-50 (32x4d), EfficinetNet-B4) and one object detection model (Yolov5) is proposed. This model classifies fire and three different colors of smoke, fog, light, and sunlight. The training of multiple as an ensemble results in their performance being superior to that of a single classification or detection model. In the classification phase, our model (FDN) achieves an F1-score that is 11.3% higher than that of the single-classification model and 31.7% higher specifically for gray smoke. In the detection phase, mAP@50 and mAP@50:95 are improved by 1.3% and 1.6%, respectively, compared to the single detection model, Yolov5s. The proposed model effectively distinguishes between fog and clouds, similar to smoke, and light and sunlight, akin to a real fire using the Partial-Loss-Balanced TaskWeighting (P-LBTW) algorithm we develop to reduce negative transfer by partially learning task-specific weights. Leveraging the classified and detected results, wildfires can be promptly and accurately identified wherever the equipment is deployed, thus mitigating the risk of large-scale fires. Therefore, the proposed real-time ensemble fire detection system offers practical models and system configurations for field deployment, facilitating early fire suppression to mitigate fire damage.