• 중소기업융합학회논문지
• /
• 제6권4호
• /
• pp.71-77
• /
• 2016

본 논문에서는 기존의 BIC(Bayesian Information Criterion) 기반 화자변화의 성능 향상을 위하여 GMM-UBM(Gaussian Mixture Model-Universal Background Model) 기반 KL(Kullback Leibler) 거리를 활용한 화자변화 검증을 제안하였다. 정보량의 차이에 민감한 기존의 BIC 기반 화자변화검출 알고리즘을 상대적으로 정보량 차이에 견인한 KL 거리 알고리즘으로 검증하였고, 정보량의 비대칭을 보상하기 위해서 GMM-UBM을 활용하였다. 기존의 BIC 기반 화자변화 검출은 1단계로 비유사도 d가 양수인 구간의 국소 최댓값인 지점을 화자변화 후보지점으로 검출하였고, 2단계로 검출된 화자변화 후보지점 중 ${\Delta}BIC$가 양수인 지점을 화자변화지점으로 결정하였다. 본 논문에서는 BIC 기반 화자변화 검출에 의해 결정된 화자변화지점에 대하여 GMM-UBM 기반 KL 거리 D가 문턱치(threshold)보다 높은 지점을 최종 화자변화 지점으로 검증하였다. 실험결과, MDR(Missed Detection Rate)이 0인 조건에서 문턱치 0.028일 때 FAR(False Alarm Rate) 60.4%로 성능이 향상되었다.

• 응용통계연구
• /
• 제30권3호
• /
• pp.377-390
• /
• 2017

설문조사를 실시할 때 응답자가 설문조사의 일부 문항에 대하여 응답하지 않는 경우 항목무응답이 발생한다. 무응답이 발생한 자료를 제외하고 완전하게 응답된 자료 만에 근거한 분석은 분석 결과에 편의가 발생할 수 있으므로, 이를 채워 넣어 완전한 형태의 자료로 분석하기 위해서 무응답 대체가 흔히 사용되고 있으며 여러 가지 무응답 대체 기법들을 비교하는 연구들도 많이 존재한다. 패널조사 연구는 연구 대상 패널에 대하여 정해진 시간에 따라 반복적으로 동일한 설문 문항에 대하여 응답을 조사하여 시간에 따른 변화를 살펴보는 조사 방법을 나타낸다. 패널조사 자료의 항목 무응답을 대체할 때 이전 시점의 응답 자료가 존재한다면 이를 포함하여 대체를 실시하는 것이 바람직한 것으로 여겨져 왔으나 이에 관한 직접적인 연구는 찾기 힘들다. 따라서 본 연구에서는 패널자료에서 이전 시점의 정보를 고려하지 않고 대체를 실시하는 방법과 이전 시점의 정보를 활용하여 대체하는 방법들 중에서 어느 대체 방법이 보다 적절한 대체를 제공하는지 살펴보았다. 특히 이전 시점의 응답 정보를 이용하는 방법인 비대체, 선형혼합모형을 이용한 대체와 선형혼합모형에 근거한 베이지안 대체 방법을 고려하였고, 이를 이전 시점의 정보를 고려하지 않는 대체 방법들 중 흔히 사용되는 평균대체, 핫덱대체 방법과 비교하였다. 모의실험 결과 선형혼합모형에 근거한 베이지 안 대체 방법이 다른 대체 방법에 비해 무응답 비율이 높아지더라도 편의도 작으며 평균에 관한 95% 신뢰구간의 포함률도 높게 나타나서 가장 좋은 대체 방법으로 확인되었다.

• Journal of Preventive Medicine and Public Health
• /
• 제51권1호
• /
• pp.33-40
• /
• 2018

Objectives: The aim of this study was to determine the factors associated with the spatial distribution of the incidence of colorectal cancer (CRC) in the neighborhoods of Tehran, Iran using Bayesian spatial models. Methods: This ecological study was implemented in Tehran on the neighborhood level. Socioeconomic variables, risk factors, and health costs were extracted from the Equity Assessment Study conducted in Tehran. The data on CRC incidence were extracted from the Iranian population-based cancer registry. The $Besag-York-Molli{\acute{e}}$ (BYM) model was used to identify factors associated with the spatial distribution of CRC incidence. The software programs OpenBUGS version 3.2.3, ArcGIS 10.3, and GeoDa were used for the analysis. Results: The Moran index was statistically significant for all the variables studied (p<0.05). The BYM model showed that having a women head of household (median standardized incidence ratio [SIR], 1.63; 95% confidence interval [CI], 1.06 to 2.53), living in a rental house (median SIR, 0.82; 95% CI, 0.71 to 0.96), not consuming milk daily (median SIR, 0.71; 95% CI, 0.55 to 0.94) and having greater household health expenditures (median SIR, 1.34; 95% CI, 1.06 to 1.68) were associated with a statistically significant elevation in the SIR of CRC. The median (interquartile range) and mean (standard deviation) values of the SIR of CRC, with the inclusion of all the variables studied in the model, were 0.57 (1.01) and 1.05 (1.31), respectively. Conclusions: Inequality was found in the spatial distribution of CRC incidence in Tehran on the neighborhood level. Paying attention to this inequality and the factors associated with it may be useful for resource allocation and developing preventive strategies in at-risk areas.

• IEIE Transactions on Smart Processing and Computing
• /
• 제2권5호
• /
• pp.277-281
• /
• 2013

Modern speaker verification systems based on support vector machines (SVMs) use Gaussian mixture model (GMM) supervectors as their input feature vectors, and the maximum a posteriori (MAP) adaptation is a conventional method for generating speaker-dependent GMMs by adapting a universal background model (UBM). MAP adaptation requires the appropriate amount of input utterance due to the number of model parameters to be estimated. On the other hand, with limited utterances, unreliable MAP adaptation can be performed, which causes adaptation noise even though the Bayesian priors used in the MAP adaptation smooth the movements between the UBM and speaker dependent GMMs. This paper proposes a sparse MAP adaptation method, which is known to perform well in the automatic speech recognition area. By introducing sparse MAP adaptation to the GMM-SVM-based speaker verification system, the adaptation noise can be mitigated effectively. The proposed method utilizes the L0 norm as a regularizer to induce sparsity. The experimental results on the TIMIT database showed that the sparse MAP-based GMM-SVM speaker verification system yields a 42.6% relative reduction in the equal error rate with few additional computations.

• 한국경영과학회지
• /
• 제12권1호
• /
• pp.19-26
• /
• 1987

In analyzing time series, estimating the level or the current mean of the process plays an important role in understanding its structure and in being able to make forecasts. The studies the class of time series models where the level of the process is assumed to follow a random walk and the deviation from the level follow an ARMA process. The estimation and forecasting problem in a Bayesian framework and uses the Kalman filter to obtain forecasts based on estimates of level. In the analysis of time series, we usually make the assumption that the time series is generated by one model. However, in many situations the time series undergoes a structural change at one point in time. For example there may be a change in the distribution of random variables or in parameter values. Another example occurs when the level of the process changes abruptly at one period. In order to study such problems, the assumption that level follows a random walk process is relaxed to include a major level change at a particular point in time. The major level change is detected by examining the likelihood raio under a null hypothesis of no change and an alternative hypothesis of a major level change. The author proposes a method for estimation the size of the level change by adding one state variable to the state space model of the original Kalman filter. Detailed theoretical and numerical results are obtained for th first order autoregressive process wirth level changes.

• 디지털융복합연구
• /
• 제12권6호
• /
• pp.325-332
• /
• 2014

일반적으로 자동분류는 학습문서의 개수에 영향을 받는다고 알려져 있지만 실제로 학습문서의 수가 텍스트 자동분류에 어떻게 영향을 주는지 입증한 연구는 거의 없었다. 본 연구에서는 학습문서 수가 자동분류에 어떤 영향을 주는지 알아보기 위해 최근에 개발된 편차기반 분류방법을 중심으로 다른 분류 알고리즘과 비교하는데 초점을 두었다. 실험결과, 편차기반 분류모델은 학습문서의 수가 총 21개(7개 장르)인 상황에서 정확도가 0.8로 베이지안이나 지지벡터기계보다 우수하게 나타났다. 이것은 편차기반 분류모델이 장르내의 주제정보를 이용하여 학습하기 때문에 학습문서의 수가 적더라도 다른 학습방법보다 좋은 자질 선택 능력을 갖는다는 것을 입증한 것이다.

• 한국GIS학회:학술대회논문집
• /
• 한국GIS학회 2003년도 공동 춘계학술대회 논문집
• /
• pp.63-63
• /
• 2003

The aim of this study is to apply and verify of Bayesian probability model, the likelihood ratio and statistical model, at Janghung, Korea, using a Geographic Information System (GIS). Landslide locations were identified in the study area from interpretation of IRS satellite images, field surveys, and maps of the topography, soil type, forest cover, geology and land use were constructed to spatial database. The factors that influence landslide occurrence, such as slope, aspect and curvature of topography were calculated from the topographic database. Texture, material, drainage and effective soil thickness were extracted from the soil database, and type, diameter and density of forest were extracted from the forest database. Land use was classified from the Landsat TM image satellite image. As each factor's ratings, the likelihood ratio coefficient were overlaid for landslide susceptibility mapping, Then the landslide susceptibility map was verified and compared using the existing landslide location. The results can be used to reduce hazards associated with landslides management and to plan land use and construction.

• 대한조선학회논문집
• /
• 제57권3호
• /
• pp.175-181
• /
• 2020

For the success of a structural integrity management, it is essential to acquire structural response data at some critical locations with limited number of sensors. In this study, the structural response of numerical model was estimated by data fusion approach based on the Kalman filter known as stochastic recursive filter. Firstly, transient direct analysis was conducted to calculate the acceleration and strain of the numerical standing beam model, then the noise signals were mixed to generate the numerical measurement signals. The acceleration measurement signal was provided to the Kalman filter as an information on the external load, and the displacement measurement, which was transformed from the strain measurement by using strain-displacement conversion relationship, was provided into the Kalman filter as an observation information. Finally, the Kalman filter estimated the displacement by combining both displacements calculated from each numerically measured signal, then the estimated results were compared with the results of the transient direct analysis.

• KDI Journal of Economic Policy
• /
• 제36권2호
• /
• pp.1-64
• /
• 2014

본고는 소규모 개방경제를 상정한 동태확률모형을 통해 2000년 이후의 한국 거시경제의 변동요인들과 금융위기 기간의 통화정책과 재정정책의 변화요인을 추정하였다. 실증분석을 위해 동태확률모형은 경제성장률을 결정하는 두 가지 다른 추세요인과 다양한 구조충격요인 그리고 다양한 형태의 마찰을 포함시킴으로써, 거시변수들이 어떠한 요인에 의해 변화했는지를 정량적으로 분석하였다. 실증분석 결과, 금융위기 기간 동안 통화정책은 주로 테일러 준칙을 따른 것으로 나타난 반면, 재정정책의 반응은 재정준칙보다는 상대적으로 큰 폭의 재정충격요인에 의해서 설명된 것으로 추정되었다.

• 대한원격탐사학회지
• /
• 제23권1호
• /
• pp.33-42
• /
• 2007

In this paper, an iterative MAP approach using a Bayesian model based on the lognormal distribution for image intensity and a GRF for image texture is proposed for despeckling the SAR images that are corrupted by multiplicative speckle noise. When the image intensity is logarithmically transformed, the speckle noise is approximately Gaussian additive noise, and it tends to a normal probability much faster than the intensity distribution. MRFs have been used to model spatially correlated and signal-dependent phenomena for SAR speckled images. The MRF is incorporated into digital image analysis by viewing pixel types as slates of molecules in a lattice-like physical system defined on a GRF Because of the MRF-SRF equivalence, the assignment of an energy function to the physical system determines its Gibbs measure, which is used to model molecular interactions. The proposed Point-Jacobian Iterative MAP estimation method was first evaluated using simulation data generated by the Monte Carlo method. The methodology was then applied to data acquired by the ESA's ERS satellite on Nonsan area of Korean Peninsula. In the extensive experiments of this study, The proposed method demonstrated the capability to relax speckle noise and estimate noise-free intensity.