• 전기전자학회논문지
• /
• 제24권4호
• /
• pp.969-974
• /
• 2020

Research into hyper parameter optimization (HPO) has recently revived with interest in models containing many hyper parameters, such as deep neural networks. In this paper, we introduce the most widely used HPO methods, such as grid search, random search, and Bayesian optimization, and investigate their characteristics through experiments. The MNIST data set is used to compare results in experiments to find the best method that can be used to achieve higher accuracy in a relatively short time simulation. The learning rate and weight decay have been chosen for this experiment because these are the commonly used parameters in this kind of experiment.

• 지구물리와물리탐사
• /
• 제23권3호
• /
• pp.157-167
• /
• 2020

최근 인공지능 기술의 발전과 함께 물리탐사의 다양한 분야에서도 인공지능의 핵심 기술인 머신러닝의 활용도가 증가하고 있다. 또한 머신러닝 및 딥러닝을 활용한 연구는 이미지, 비디오, 음성, 자연어 등 다양한 태스크의 추론 정확도를 높이기 위해 복잡한 알고리즘들이 개발되고 있고, 더 나아가 자료의 특성, 알고리즘 구조 및 하이퍼 파라미터의 최적화를 위한 자동 머신러닝(AutoML) 분야로 그 폭을 넓혀가고 있다. 본 연구에서는 AutoML 분야 중에서도 하이퍼 파라미터(hyperparameter) 자동 탐색을 위한 베이지안 최적화 기술에 중점을 두었으며, 본 기술을 물리탐사 분야에서도 암상 분류(facies classification) 문제에 적용했다. Vincent field의 현장 물리검층 및 탄성파 자료를 이용하여 암상 및 공극유체를 분류하는 지도학습 기반 모델에 적용하였고, 랜덤 탐색 기법의 결과와 비교하여 베이지안 최적화 기반 예측 프레임워크의 효율성을 검증하였다.

• 품질경영학회지
• /
• 제22권1호
• /
• pp.96-112
• /
• 1994

This article studies the design of Bayesian single attribute acceptance sampling plans under dependent production processes. An economic model is constructed by extending the mathematical model developed for non-Bayesian cases for Bayesian cases. The mathematical structure of the model is analyzed and it is used to prove that optimization of the model can be achieved by applying the solution method developed for non-Bayesian models directly. The effect of dependence patterns and the types of prior distributions on the design of sampling plans is also investigated through a computational study.

• 한국정보과학회논문지:소프트웨어및응용
• /
• 제36권3호
• /
• pp.208-216
• /
• 2009

유전자알고리즘의 교차나 돌연변이 연산을 직접적으로 사용하지 않고 개체군의 확률분포를 추정하여 보다 효율적인 탐색을 수행하려는 분포추정알고리즘이 여러 방법으로 제안되었다. 그러나 실제로 변수들간의 고차상관관계를 파악하는 일은 쉽지 않은 일이라 대부분의 경우 낮은 차수의 상관관계를 제한된 가정하에 추정하게 된다. 본 논문에서는 데이타의 고차상관관계를 표현할 수 있고 최적 해를 좀 더 효율적으로 찾을 수 있는 새로운 분포추정알고리즘을 제안한다. 제안된 알고리즘에서는 상관관계가 있을 것으로 추정되는 변수들의 집합으로 정의된 하이퍼에지로 구성된 랜덤 하이퍼그래프 모델을 구축하여 변수들 간의 고차상관관계를 표현하고, 베이지안 샘플링 알고리즘(Bayesian Sampling Algorithm)을 통해 다음 세대의 개체를 생성한다. 기만하는 빌딩블럭(deceptive building blocks)을 가진 분해가능(decomposable) 함수에 대하여 실험한 결과 성공적으로 최적해를 구할 수 있었으며 단순 유전자알고리즘과 BOA (Bayesian Optimization Algorithm)와 비교하여 좋은 성능을 얻을 수 있었다.

• 디지털융복합연구
• /
• 제12권7호
• /
• pp.273-278
• /
• 2014

HMM(Hidden Markov Model)을 이용한 어휘 인식에서 인식 어휘의 모델들의 대한 인식 확률이 이산적인 분포를 나타내며 인식을 위한 계산량이 적은 장점이 있지만 인식률을 계산했을 때 상대적으로 낮은 단점이 있다. 이를 개선하기 위하여 HMM(Hidden Markov Model) 모델 최적화를 이용한 베이시안 기법 인식률 향상을 제안한다. 본 논문은 HMM 어휘 인식에서 인식을 위한 모델 구성을 가우시안 믹스쳐 모델로 최적화한 인식 모델을 생성하였으며 베이시안 기법인 사전확률과 사후확률을 이용한 인식률을 향상시켰다. 본 논문에서 제안한 방법을 적용한 결과 어휘인식률에서 97.9%의 인식률을 나타내었다.

• 한국방재학회:학술대회논문집
• /
• 한국방재학회 2008년도 정기총회 및 학술발표대회
• /
• pp.723-726
• /
• 2008

It is not always easy to estimate the parameters in hydrologic models due to insufficient hydrologic data when hydraulic structures are designed or water resources plan are established, uncertainty analysis, therefore, are inevitably needed to examine reliability for the estimated results. With regard to this point, this study applies a Bayesian Markov Chain Monte Carlo scheme to the NWS-PC rainfall-runoff model that has been widely used, and a case study is performed in Soyang Dam watershed in Korea. The NWS-PC model is calibrated against observed daily runoff, and thirteen parameters in the model are optimized as well as posterior distributions associated with each parameter are derived. The Bayesian Markov Chain Monte Carlo shows a improved result in terms of statistical performance measures and graphical examination. The patterns of runoff can be influenced by various factors and the Bayesian approaches are capable of translating the uncertainties into parameter uncertainties. One could provide against an expected runoff event by utilizing information driven by Bayesian methods. Therefore, the rainfall-runoff analysis coupled with the uncertainty analysis can give us an insight in evaluating flood risk and dam size in a reasonable way.

• Nuclear Engineering and Technology
• /
• 제54권4호
• /
• pp.1363-1373
• /
• 2022

The recent seismic events that occurred in South Korea have increased the interest in the re-evaluation of the seismic capacity of nuclear power plant (NPP) equipment, which is often conservatively estimated. To date, various approaches-including the Bayesian method proposed by the United States (US) Electric Power Research Institute -have been developed to quantify the seismic capacity of NPP equipment. Among these, the Bayesian approach has advantages in accounting for both prior knowledge and new information to update the probabilistic distribution of seismic capacity. However, data availability and region-specific issues exist in applying this Bayesian approach to Korean NPP equipment. Therefore, this paper proposes to construct an earthquake experience database by combining available earthquake records at Korean NPP sites and the general location of equipment within NPPs. Also, for the better representation of the seismic demand of Korean earthquake datasets, which have distinct seismic characteristics from those of the US at a high-frequency range, a broadband frequency range optimization is suggested. The proposed data construction and seismic demand optimization method for seismic capacity re-evaluation are demonstrated and tested on a 480 V motor control center of a South Korea NPP.

• Smart Structures and Systems
• /
• 제33권1호
• /
• pp.17-26
• /
• 2024

The new metro crossing the existing metro will cause the settlement or floating of the existing structures, which will have safety problems for the operation of the existing metro and the construction of the new metro. Therefore, it is necessary to monitor and predict the settlement of the existing metro caused by the construction of the new metro in real time. Considering the complexity and uncertainty of metro settlement, a Gaussian Prior Bayesian Emulator (GPBE) probability prediction model based on Bayesian optimization (BO) is proposed, that is, BO-GPBE. Firstly, the settlement monitoring data are analyzed to get the influence of the new metro on the settlement of the existing metro. Then, five different acquisition functions, that is, expected improvement (EI), expected improvement per second (EIPS), expected improvement per second plus (EIPSP), lower confidence bound (LCB), probability of improvement (PI) are selected to construct BO model, and then BO-GPBE model is established. Finally, three years settlement monitoring data were collected by structural health monitoring (SHM) system installed on Nanjing Metro Line 10 are employed to demonstrate the effectiveness of BO-GPBE for forecasting the settlement.

• 한국전산구조공학회:학술대회논문집
• /
• 한국전산구조공학회 2008년도 정기 학술대회
• /
• pp.602-607
• /
• 2008

A technique for reliability-based design optimization(RBDO) is developed based on the Bayesian approach, which can deal with the epistemic uncertainty arising due to the limited number of data. Until recently, the conventional RBDO was implemented mostly by assuming the uncertainty as aleatory which means the statistical properties are completely known. In practice, however, this is not the case due to the insufficient data for estimating the statistical information, which makes the existing RBDO methods less useful. In this study, a Bayesian reliability is introduced to take account of the epistemic uncertainty, which is defined as the lower confidence bound of the probability distribution of the original reliability. In this case, the Bayesian reliability requires double loop of the conventional reliability analyses, which can be computationally expensive. Kriging based dimension reduction method(KDRM), which is a new efficient tool for the reliability analysis, is employed to this end. The proposed method is illustrated using a couple of numerical examples.

• Smart Structures and Systems
• /
• 제21권4호
• /
• pp.435-448
• /
• 2018

This paper reports the development of a theoretically rigorous method for permanent way engineers to assess the condition of railway ballast under a concrete sleeper with the potential to be extended to a smart system for long-term health monitoring of railway ballast. Owing to the uncertainties induced by the problems of modeling error and measurement noise, the Bayesian approach was followed in the development. After the selection of the most plausible model class for describing the damage status of the rail-sleeper-ballast system, Bayesian model updating is adopted to calculate the posterior PDF of the ballast stiffness at various regions under the sleeper. An obvious drop in ballast stiffness at a region under the sleeper is an evidence of ballast damage. In model updating, the model that can minimize the discrepancy between the measured and model-predicted modal parameters can be considered as the most probable model for calculating the posterior PDF under the Bayesian framework. To address the problems of non-uniqueness and local minima in the model updating process, a two-stage hybrid optimization method was developed. The modified evolutionary algorithm was developed in the first stage to identify the important regions in the parameter space and resulting in a set of initial trials for deterministic optimization to locate all most probable models in the second stage. The proposed methodology was numerically and experimentally verified. Using the identified model, a series of comprehensive numerical case studies was carried out to investigate the effects of data quantity and quality on the results of ballast damage detection. Difficulties to be overcome before the proposed method can be extended to a long-term ballast monitoring system are discussed in the conclusion.