• 산업경영시스템학회지
• /
• 제43권4호
• /
• pp.107-115
• /
• 2020

Support vector regression (SVR) is devised to solve the regression problem by utilizing the excellent predictive power of Support Vector Machine. In particular, the ⲉ-insensitive loss function, which is a loss function often used in SVR, is a function thatdoes not generate penalties if the difference between the actual value and the estimated regression curve is within ⲉ. In most studies, the ⲉ-insensitive loss function is used symmetrically, and it is of interest to determine the value of ⲉ. In SVQR (Support Vector Quantile Regression), the asymmetry of the width of ⲉ and the slope of the penalty was controlled using the parameter p. However, the slope of the penalty is fixed according to the p value that determines the asymmetry of ⲉ. In this study, a new ε-insensitive loss function with p1 and p2 parameters was proposed. A new asymmetric SVR called GSVQR (Generalized Support Vector Quantile Regression) based on the new ε-insensitive loss function can control the asymmetry of the width of ⲉ and the slope of the penalty using the parameters p1 and p2, respectively. Moreover, the figures show that the asymmetry of the width of ⲉ and the slope of the penalty is controlled. Finally, through an experiment on a function, the accuracy of the existing symmetric Soft Margin, asymmetric SVQR, and asymmetric GSVQR was examined, and the characteristics of each were shown through figures.

• 대한수학회지
• /
• 제59권2호
• /
• pp.217-233
• /
• 2022

Although machine learning shows state-of-the-art performance in a variety of fields, it is short a theoretical understanding of how machine learning works. Recently, theoretical approaches are actively being studied, and there are results for one of them, margin and its distribution. In this paper, especially we focused on the role of margin in the perturbations of inputs and parameters. We show a generalization bound for two cases, a linear model for binary classification and neural networks for multi-classification, when the inputs have normal distributed random noises. The additional generalization term caused by random noises is related to margin and exponentially inversely proportional to the noise level for binary classification. And in neural networks, the additional generalization term depends on (input dimension) × (norms of input and weights). For these results, we used the PAC-Bayesian framework. This paper is considering random noises and margin together, and it will be helpful to a better understanding of model sensitivity and the construction of robust generalization.

• Nuclear Engineering and Technology
• /
• 제48권3호
• /
• pp.684-701
• /
• 2016

The Gaussian process model (GPM) is a flexible surrogate model that can be used for nonparametric regression for multivariate problems. A unique feature of the GPM is that a prediction variance is automatically provided with the regression function. In this paper, we estimate the safety margin of a nuclear power plant by performing regression on the output of best-estimate simulations of a large-break loss-of-coolant accident with sampling of safety system configuration, sequence timing, technical specifications, and thermal hydraulic parameter uncertainties. The key aspect of our approach is that the GPM regression is only performed on the dominant input variables, the safety injection flow rate and the delay time for AC powered pumps to start representing sequence timing uncertainty, providing a predictive model for the peak clad temperature during a reflood phase. Other uncertainties are interpreted as contributors to the measurement noise of the code output and are implicitly treated in the GPM in the noise variance term, providing local uncertainty bounds for the peak clad temperature. We discuss the applicability of the foregoing method to reduce the use of conservative assumptions in best estimate plus uncertainty (BEPU) and Level 1 probabilistic safety assessment (PSA) success criteria definitions while dealing with a large number of uncertainties.

• 한국원자력학회:학술대회논문집
• /
• 한국원자력학회 1995년도 추계학술발표회논문집(2)
• /
• pp.562-567
• /
• 1995

The concept of VLDT (Variable Low DNBR Trip), a new CPC trip function, was proposed and applied to the events of increase in secondary heat removal, such as an excess feedwater event anti an IOSGADV (Inadvertent Opening S/G Atmospheric Dump Valve). Major assumption used in this study was no time delay to LOOP (Loss of Offsite Power) after turbine trip. In case of using this VLDT function, safety criterion of DNB would not be violated under the same condition as previous analysis without any change in thermal margin.

• 대한안전경영과학회지
• /
• 제13권1호
• /
• pp.203-209
• /
• 2011

This study develops three new models that are practically applicable to three stages of Taguchi's robust design, which includes system design, parameter design and tolerance design. In system design, the Multiple Loss Function Analysis(MLFA) and Overall Loss Index(OLI) which reflect upon weight of characteristics and importance of specification are developed. Moreover parameter design presents Process Capability Index(PCI), $C_{PUK}$ and $C_{PLK}$, in order to segregate Signal-To-Noise Ratio(SNR) into accuracy and precision for an evaluation of relative comparison. In addition, tolerance design presents the new model of allowance computation for assembled product which is primarily derived from safety margin(SM) considering functional limit and specification. The guideline of those three new models, which include systematic charts and applicable illustrations, offers convenience for practitioners in the field. Hence, the practical applications could be made with the steps of robust designs such as system design, parameter design and specification allowance design.

• 한국융합학회논문지
• /
• 제12권3호
• /
• pp.17-25
• /
• 2021

의복을 착용하는데 있어 목적 상황에 부합하는 옷차림을 구성하는 것은 중요하다. 따라서 인공지능 기반의 다양한 패션 추천 시스템에서 의복 착용의 T.P.O(Time, Place, Occasion)를 고려하고 있다. 하지만 옷차림으로부터 직접 T.P.O를 추론하는 연구는 많지 않은데, 이는 문제 특성 상 다중 레이블 및 클래스 불균형 문제가 발생하여 모델 학습을 어렵게 하기 때문이다. 이에 본 연구에서는 label-distribution-aware margin(LDAM) loss를 도입하여 옷차림의 T.P.O를 추론할 수 있는 모델을 제안한다. 모델의 학습 및 평가를 위한 데이터셋은 패션 쇼핑몰로부터 수집되었고 이를 바탕으로 성능을 측정한 결과, 제안 모델은 비교 모델 대비 모든 T.P.O 클래스에서 균형잡힌 성능을 보여주는 것을 확인할 수 있었다.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 1996년도 하계학술대회 논문집 B
• /
• pp.1018-1020
• /
• 1996

In this paper, when a robust active noise controller for a small cavity to control the noise induced in the cavity is designed, the Graphical method based on the robust stability and performance requirements is studied. The problem of designing controller that achieve these robust performance conditions is related to minimizing the $H_{\infty}$ norm of the mixed sensitivity function by using $H_{\infty}$ control theory. Also, For design the controller, the loopshaping method which control the weight functions to satisfy the design specification without loss of a robust performance can be used. Therefore, we determined the acceptable design specification with the system characteristics of the small cavity and obtained its robust controller with the robust performance specifications by stability margin.

• 한국방송∙미디어공학회:학술대회논문집
• /
• 한국방송∙미디어공학회 2022년도 추계학술대회
• /
• pp.171-174
• /
• 2022

Weakly-supervised learning is a widely adopted approach in video anomaly detection whereby only video labels are utilized instead of expensive frame-level annotations. Since the success of multi-instance learning (MIL), almost all recent approaches are based on maximizing the margin between the set of abnormal video snippets and those of normal video snippets. In this work, we present a simple contrastive approach for weakly supervised video anomaly detection (WS-VAD) with aims to enhance the performance of existing models. The method is generic in nature and introduces a loss function to encourage attraction of output features from the same video class and repel those from different video classes. Experimental results demonstrate our method can be applied to existing algorithms to improve detection accuracy in public video anomaly dataset.

• 한국방송∙미디어공학회:학술대회논문집
• /
• 한국방송∙미디어공학회 2022년도 하계학술대회
• /
• pp.224-227
• /
• 2022

객체 분류는 입력으로 주어진 이미지에 포함된 객체의 종류를 판단하는 기술이다. 대표적인 딥러닝 기반의 객체 분류 방법으로서 Faster R-CNN[2], YOLO[3] 등의 모델이 개발되었으나, 여전히 성능 향상의 여지가 있다. 본 연구에서는 각도 마진 손실 함수를 기존의 몇 가지 객채 분류 모델에 적용하여 성능 향상을 유도한다. 각도 마진 손실 함수는 얼굴 인식 모델인 SphereFace [4]에서 제안한 방법으로, 얼굴 인식과 같이 단일 도메인의 데이터셋을 분류하는 문제를 풀기 위해 제안되었다. 이는 기존 소프트맥스 함수에서 클래스 결정 경계선에 마진을 주는 방식으로 클래스 간의 구분 능력을 향상시킨다. 본 논문은 각도 마진 손실 함수를 CIFAR10, CIFAR100 데이터셋의 분류 문제에 적용하였으며 ResNet, EfficientNet, MobileNet 등의 백본 네트워크로 실험하여 평균적으로 mAP 성능이 향상되는 것을 확인하였다.

• Journal of Electrical Engineering and Technology
• /
• 제5권3호
• /
• pp.353-362
• /
• 2010

This paper presents a centralized control algorithm for power system performance in the Korean power system using Flexible AC Transmission Systems (FACTS) devices. The algorithm is applied to the Korean power system throughout the metropolitan area in order to alleviate inherent stability problems, especially concerns with voltage stability. Generally, control strategies are divided into local and centralized control. This paper is concerned with a centralized control strategy in terms of the global system. In this research, input data of the proposed algorithm and network data are obtained from the SCADA/EMS system. Using the full system model, the centralized controller monitors the system condition and decides the operating point according to the control objectives that are, in turn, dependent on system conditions. To overcome voltage collapse problems, load-shedding is currently applied in the Korean power system. In this study, the application of the coordination between FACTS and switch capacitor (SC) can restore the solvability without load shedding or guarantee the FV margin when the margin is insufficient. Optimal Power Flow (OPF) algorithm, for which the objective function is loss minimization, is used in a stable case. The results illustrate examples of the proposed algorithm using SCADA/EMS data of the Korean power system in 2007.