Heecheon You Applied Sciences-basel

[논문] Machine Learning-Based Smartphone Grip Posture Image Recognition and Classification

Abstract: Uncomfortable smartphone grip postures resulting from inappropriate user inter face design can degrade smartphone usability. This study aims to develop a classification model for smartphone grip postures by detecting the positions of the hand and fingers on smartphones using machine learning techniques. Seventy participants (35 males and 35 females with an average of 38.5 ± 12.2 years) with varying hand sizes participated in the smartphone grip posture experiment. The participants performed four tasks (making calls, listening to music, sending text messages, and web browsing) using nine smartphone mock ups of different sizes, while cameras positioned above and below their hands recorded their usage. A total of 3278 grip posture images were extracted from the recorded videos and were preprocessed using a skin color and hand contour detection model. The grip postures were categorized into seven types, and three models (MobileNetV2, Inception V3, and ResNet-50), along with an ensemble model, were used for classification. The ensemble-based classification model achieved an accuracy of 95.9%, demonstrating higher accuracy than the individual models: MobileNetV2 (90.6%), ResNet-50 (94.2%), and Incep tion V3 (85.9%). The classification model developed in this study can efficiently analyze grip postures, thereby improving usability in the development of smartphones and other electronic devices.

Kwang-Jae Kim Expert Systems with Applications

[논문] Development of a suspiciousness measure with reduced redundancy for screening machine sets in serial-parallel multistage manufacturing processes

ABSTRACT In serial–parallel multistage manufacturing processes (SP-MMPs), identifying root cause machine sets (MSs) that contribute to defective product production is crucial for improving product quality. To facilitate the efficient identification of root cause MSs, suspicious MSs are typically screened using association rule-based approaches. Aredundant MS contains a root cause MS but does not provide any additional information beyond what the root cause MS already offers. Conventional suspicious MS screening measures calculate suspiciousness based on machine assignment history. However, these measures often overrate redundant MSs because the association of the root cause MS with defective product production is also considered when calculating the suspiciousness of the redundant MS. This overrating can result in redundant MSs being prioritized during screening, while actual root cause MSs are overlooked. This study argues that a suspiciousness measure should assess the association between an MS and defective product production, excluding associations that can be explained by any proper subset MSs of the MS. To address this, a novel suspiciousness measure based on multi-item association is proposed. Case studies conducted on simulated SP-MMPs demonstrate that the proposed measure effectively reduces the overrating of redundant MSs and improves the prioritization of root cause MSs during the screening process. These findings demonstrate the effectiveness of the proposed measure in suspicious MS screening

Kwangmin Jung Geneva Papers on Risk and Insurance: Issues and Practice

[논문] The effect of corporate risk management on cyber risk mitigation: Evidence from the insurance industry

Abstract We examine how corporate risk management can be used to address a firm’s vulner ability to cyber risk. We use a large, novel dataset on cyber risk and corporate risk management to analyse US insurers’ cyber loss events during the period of 2000-2021. Our analysis includes information on whether insurers have implemented an enterprise risk management (ERM) programme and whether they report applying cyber risk management (CRM). The results illustrate that the implementation of CRM measures may have no significant effect on cyber risk mitigation. However, we determine that the likelihood (frequency) of a cyber loss event decreases by 3.9% (6.8%) as ERM programmes mature year on year. We also find that an insurer can benefit from implementing both CRM and ERM through a lowered event likelihood (frequency) of 3.8 percentage points on average (3.7 percentage points) per year compared to solely implementing an ERM programme

Kwangmin Jung 보험학회지

[논문] 시스템적 사이버 리스크 분류체계 및 통계적 특성 연구

Abstract 시스템적 사이버 리스크는 기업 및 산업 전반에 막대한 경제적 손실을 초래할 수 있는 위협으로 부각되고 있다. 본 연구는 이러한 리스크의 핵심 요인을 정의하고 손실 데이터를 추출하여 업종별 기업의 재정 부담 수준 평가를 목적으로 한다. 이를 위해 텍스트 마이닝 기법을 활용하여 시스템적 사이버 손실 사건을 식별하고, 월별 빈도와 개별 심도의 분포 적합성 검정을 수행하였으며, 손실분포접근법을 통해 총손실분포를 추정하였다. 분석 결과, 빈도는 음이항 분포를, 심도는 로그 정규 분포와 일반화 파레토 분포의 결합 분포를 따르는 것을 확인하였다. 이러한 시스템적 사이버 손실에 대해 제조 및 정보 산업이 상대적으로 높은 재정 부담을 지는 반면, 금융 및 소매업의 부담은 낮은 것으로 평가하였다. 본 연구 결과를 통해 업종별 차별화된 리스크 평가 및 요율 산정의 중요성을 강조함으로써 글로벌 사이버 보험시장에서의 공급 역량을 강화하는데 시사점을 제공할 수 있다.

Young Myoung Ko Technometrics

[논문] Strata Design for Variance Reduction in Stochastic Simulation

Abstract Stratified sampling is one of the powerful variance reduction methods for analyzingsystem performance, such as reliability, with stochastic simulation. It divides the inputspace into disjoint subsets, called strata, to draw samples from each stratum.Partitioning the input space properly and allocating greater computational effort tocrucial strata can help accurately estimate system performance with a limitedcomputational budget. How to create strata, however, has yet to be thoroughlyexamined. Strata design faces the curse of dimensionality and data scarcity as theinput dimension increases. We analytically derive the optimal stratification structurethat minimizes the estimation variance for univariate problems. Further, reconciling the optimal stratification into decision trees, we devise a robust algorithm for multidimensional problems. Numerical experiments and a wind turbine case study demonstrate the superiority of the proposed method in terms of variance reduction, leading to computational efficiency and scalability.

Bong-Gyu Jang 선물연구

[논문] High-dimensional parameter calibration of interest rate model for the Korean insurance capital standard

Abstract We propose a method for calibrating high-dimensional parameters in the Hull–White one-factor model using market prices of swaptions, aimed at generating mark-to-market interest rate scenarios in the Korean insurance industry. Our approach integrates a trust region-based Bayesian optimization technique with a parameter space decomposition method to solve the calibration problem. Empirical studies demonstrate that our method achieves superior stability and effectiveness in calibrating high-dimensional parameters compared to conventional Bayesian optimization approaches.

Young Myoung Ko/Minwoo Chae Journal of the Korean Statistical Society

[논문] Using statistical models for optimal packaging in semiconductor manufacturing processes

Abstract The importance of the back-end process in semiconductor manufacturing has recently received significant attention from global manufacturers. The analysis of manufacturing data often provides crucial insights into problems inherent in the manufacturing processes. An important goal of the back-end process is to improve the yield of final products, called packages. A simple way to achieve this goal is to characterize low-quality wafers based on the analysis of manufacturing data and discard them before proceeding to the packaging step. Alternatively, this paper proposes a novel packaging method that significantly improves the package yield using statistical models scoring the quality of dies. We prove that the proposed packaging method is optimal and conduct thorough numerical experiments, showing its superiority.

Kangbok Lee European Journal of Operational Research

[논문] The circular balancing problem

Abstract We propose a balancing problem with a minmax objective in a circular setting. This balancing problem involves the arrangement of an even number of items with different weights on a circle while minimizing the maximum total weight of items arranged on any half circle. Due to its generic structure, it may have applications in fair resource allocation schemes. We show the NP-hardness of the problem and develop polynomial-time algorithms when the number of distinct weights is a fixed constant. We propose for the general case a tight 7∕6-approximation algorithm and show that it performs better than two existing algorithms designed for an equivalent problem in the literature. The worst-case performance ratio is derived through a linear combination of valid inequalities that are obtained from the problem definition, the properties of the proposed algorithm, and the optimal circular permutation structure. Furthermore, we formulate a more general problem of minimizing the maximum total weight of items on equally divided circular sectors and present its computational complexity and a tight approximation algorithm.

Heecheon You Ergonomics

[논문] Optimising computer vision-based ergonomic assessments: sensitivity to camera position and monocular 3D pose model

Abstract Numerous computer vision algorithms have been developed to automate postureanalysis and enhance the efficiency and accuracy of ergonomic evaluations. However,the most effective algorithm for conducting ergonomic assessments remainsuncertain. Therefore, the aim of this study was to identify the optimal camera positionand monocular 3D pose model that would facilitate precise and efficient ergonomicevaluations. We evaluated and compared four currently available computer vision algorithms: Mediapipe BlazePose, VideoPose3D, 3D-pose-baseline, and PSTMO todetermine the most suitable model for conducting ergonomic assessments. Based onthe findings, the side camera position yielded the lowest Mean Absolute Error (MAE)across static, dynamic, and combined tasks. This positioning proved to be the mostreliable for ergonomic assessments. Additionally, VP3D_FB demonstrated superiorperformance among evaluated models.

Heecheon You Applied Sciences-basel

[논문] G-UNETR++: A Gradient-Enhanced Network for Accurate and Robust Liver Segmentation from Computed Tomography Images

Abstract Accurate liver segmentation from computed tomography (CT) scans is essential for liver cancer diagnosis and liver surgery planning. Convolutional neural network (CNN)-based models have limited segmentation performance due to their localized receptive fields. Hybrid models incorporating CNNs and transformers that can capture long-range dependencies have shown promising performance in liver segmentation with the cost of high model complexity. Therefore, a new network architecture named G-UNETR++ is proposed to improve accuracy in liver segmentation with moderate model complexity. Two gradientbased encoders that take the second-order partial derivatives (the first two elements from the last column of the Hessian matrix of a CT scan) as inputs are proposed to learn the 3D geometric features such as the boundaries between different organs and tissues. In addition, a hybrid loss function that combines dice loss, cross-entropy loss, and Hausdorff distance loss is designed to address class imbalance and improve segmentation performance in challenging cases. The proposed method was evaluated on three public datasets, the Liver Tumor Segmentation (LiTS) dataset, the 3D Image Reconstruction for Comparison of Algorithms Database (3D-IRCADb), and the Segmentation of the Liver Competition 2007 (Sliver07) dataset, and achieved 97.38%, 97.50%, and 97.32% in terms of the dice similarity coefficient for liver segmentation on the three datasets, respectively. The proposed method outperformed the other state-of-the-art models on the three datasets, which demonstrated the strong effectiveness, robustness, and generalizability of the proposed method in liver segmentation.