• 대한음성학회:학술대회논문집
• /
• 대한음성학회 2006년도 추계학술대회 발표논문집
• /
• pp.24-27
• /
• 2006

This paper describes how a domain dependent pronunciation lexicon is generated and optimized for Korean large vocabulary continuous speech recognition(LVCSR). At the level of lexicon, pronunciation variations are usually modeled by adding pronunciation variants to the lexicon. We propose the criteria for selecting appropriate pronunciation variants in lexicon: (i) likelihood and (ii) frequency factors to select variants. Our experiment is conducted in three steps. First, the variants are generated with knowledge-based rules. Second, we generate a domain dependent lexicon which includes various numbers of pronunciation variants based on the proposed criteria. Finally, the WERs and RTFs are examined with each lexicon. In the experiment, 0.72% WER reduction is obtained by introducing the variants pruning criteria. Furthermore, RTF is not deteriorated although the average number of variants is higher than that of compared lexica.

• Journal of Communications and Networks
• /
• 제9권4호
• /
• pp.438-445
• /
• 2007

In this paper, we consider the problem of optimum selection of quality-of-service(QoS) classes in interworking between the networks in a next-generation-network(NGN) environment. After introducing the delay-cost and loss-cost characteristics, we discuss the time-invariant(TI) and time-variant(TV) scenarios. For the TI case, we show that under nearly lossless transmission condition, each network can make its own optimization regardless of other networks. For the TV case, we present sufficient conditions under which the optimum QoS class of each network can be considered fixed with respect to time without considerable degradation in the optimization target. Therefore, under the conditions presented in this paper, the QoS of a flow in each network can be determined solely by considering the characteristics of that network and this QoS class can be held fixed during the flow period.

• 응용통계연구
• /
• 제5권2호
• /
• pp.181-192
• /
• 1992

본 논문에서는 텐서 스플라인을 이용하여, 일반화된 선형모형의 회귀합수를 자료에만 의존 하는 방식으로 추정하는 문제를 고려하였다. 최우 추정법을 이용하여 회귀 함수를 추정하는 데, 이용된 텐서 스틀라인은 접목점의 수가 유한개이며, 독립변수 영역의 주변에서는 선형으 로 제한되었다. 접목점을 자료의 각 좌표의 순서 통계량에 위치하도록 했고 그 수는 AIC의 변형된 식을 최소로 하는 수로 결정 했다. 모의 실험 예를 통하여 추정량을 예시하였다.

• Communications for Statistical Applications and Methods
• /
• 제26권2호
• /
• pp.103-129
• /
• 2019

The main goal of a case-control study is to learn the association between various risk factors and a primary outcome (e.g., disease status). Particularly recently, it is also quite common to perform secondary analyses of the case-control data in order to understand certain associations between the risk factors of the primary outcome. It has been repeatedly documented with case-control data, association studies of the risk factors that ignore the case-control sampling scheme can produce highly biased estimates of the population effects. In this article, we review the issues of the naive secondary analyses that do not account for the biased sampling scheme, and also the various methods that have been proposed to account for the case-control ascertainment. We additionally compare the results of many of the discussed methods in an example examining the association of a particular genetic variant with smoking behavior, where the data were obtained from a lung cancer case-control study.

• 한국컴퓨터정보학회논문지
• /
• 제18권4호
• /
• pp.19-26
• /
• 2013

의사 샘플 신경망은 학습 샘플의 수가 적은 경우 학습된 신경망이 국부 최적해에 빠져 성능이 저하되는 것을 보완하기 위해 기존 샘플들로부터 의사 샘플을 생성하고 이를 통해 해공간을 평탄화 시킴으로써 학습된 신경망의 성능을 향상시킬 수 있는 신경망의 변형이다. 이는 학습 샘플의 양에 관한 문제로 이 논문에서는 이에 더해 학습 샘플의 질을 향상시킴으로써 학습된 신경망의 성능을 더욱 높일 수 있는 방법을 제시하였다. 잡음이 적게 포함된 전형적인 학습 샘플들만이 주어지고 입력 특징 중 출력과 연관성이 높은 특징만을 사용함으로써 학습된 신경망의 성능을 높일 수 있음은 자명하다. 따라서 이 논문에서는 커널밀도 추정을 통해 비전형적인 학습샘플을 제거하고 입력값이 출력값에 미치는 영향을 나타내는 연관성 척도를 사용하여 연관성이 적은 특징을 제거함으로써 의사 샘플 신경망의 성능을 향상시킬 수 있음을 보였다. 제시한 방법의 유효성은 토석류 데이터를 이용한 실험을 통해 확인할 수 있다.

• 응용통계연구
• /
• 제37권1호
• /
• pp.103-118
• /
• 2024

Horseshoe 사전분포는 희박 회귀 분석에서 가장 자주 사용되는 사전분포 중 하나이다. Horseshoe 사전분포는 탐험해야 할 모수공간이 spike and slab 사전분포보다 작다는 장점이 있으나 MCMC 단계 당 시간이 오래 걸린다는 단점이 있다. 이로 인해 horseshoe 사전분포를 고차원 데이터 분석으로 확장하는 것은 한계가 존재한다. 이러한 한계를 극복하기 위해 horseshoe 사전분포에 대한 MCMC 알고리듬들이 많이 제안되었으며, 특히 최근 Johndrow 등 (2020)은 빠른 계산을 위한 근사 알고리듬을 제안하였다. 본 논문에서는 horseshoe 사전분포에 대한 (1) 전통적인 MCMC 알고리듬, (2) 근사 알고리듬 그리고 (3) 근사 알고리듬의 변형 알고리듬의 성능을 다양한 모의실험을 통해 비교하려 한다. 성능은 계산 시간, 회귀계수 추정 및 변수선택을 중심으로 비교한다. 이때 변수선택을 위해 Li와 Pati (2017)에서 제안한 horseshoe 사전분포 기반 변수선택법을 고려한다.

• Archives of Plastic Surgery
• /
• 제40권2호
• /
• pp.91-96
• /
• 2013

Background A thinned anterolateral thigh (ALT) flap is often harvested to achieve optimal skin resurfacing. Several techniques have been described to thin an ALT flap including an adipocutaneous flap, an adipofascial flap and delayed debulking. Methods By systematically reviewing all of the available literature in English and French, the present manuscript attempts to identify the common surgical indications, complications and donor site morbidity of the adipofascial variant of the ALT flap. The studies were identified by performing a systematic search on Medline, Ovid, EMBASE, the Cochrane Database of Systematic Reviews, Current Contents, PubMed, Google, and Google Scholar. Results The study selection process was adapted from the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement, and 15 articles were identified using the study inclusion criteria. These articles were then reviewed for author name(s), year of publication, flap dimensions and thickness following defatting, perforator type, type of transfer, complications, thinning technique, number of cases with a particular area of application and donor site morbidity. Conclusions The adipofascial variant of the ALT flap provides tissue to fill large defects and improve pliability. Its strong and safe blood supply permits adequate immediate or delayed debulking without vascular complications. The presence of the deep fascia makes it possible to prevent sagging by suspending and fixing the flap for functional reconstructive purposes (e.g., the intraoral cavity). Donor site morbidity is minimal, and thigh deformities can be reduced through immediate direct closure or liposuction and direct closure. A safe blood supply was confirmed by the rate of secondary flap debulking.

• Journal of Ginseng Research
• /
• 제39권2호
• /
• pp.94-104
• /
• 2015

Background: Ginseng (Panax ginseng Meyer) is an important medicinal herbs in Asia. However, ginseng varieties are less developed. Method: To developed ginseng varieties, a pure line selection method was applied in this study. Results: Gumpoong was testing of 4-yr-old specimens in 2002, the proportions of the below-ground roots that were rusty colored for Gumpoong was 1.29 in Daejeon and 1.45 in Eumseong, whereas the proportions for its yellow berry variant were 2.60 and 2.45 in the two regions, respectively. Thus the Gumpoong was resistant to root rust. Sunpoong has a high yielding property. Its average root weight is 70.6 g for 6-yr-old roots. Its yield is 2.9 kg/$1.62m^2$ and the rate of heaven- and earth-grade product is 20.9%, which is very high compared to 9.4% for Yunpoong. Sunone is resistance to root rot and the survival rate of 4-yr-old roots was 44.4% in 1997, whereas that of the violet-stem variant landrace was 21.7%. Sunhyang has content of arginyl-fructosyl-glucose (AFG), which produces the unique scent of red ginseng, is $95.1{\mu}mol/g$ and greater than the $30.8{\mu}mol/g$ of Chunpoong in 6-yr-old plants. Sunun and Cheongsun are being nurtured to protect genetic resources. Conclusion: Developed ginsneg varieties will be used as the basis for the protection of genetic resources and breeding.

• Animal Bioscience
• /
• 제37권7호
• /
• pp.1168-1176
• /
• 2024

Objective: As a charismatic species, cashmere goats have rich genetic resources. In the Inner Mongolia Autonomous Region, there are three cashmere goat varieties named and approved by the state. These goats are renowned for their high cashmere production and superior cashmere quality. Therefore, it is vitally important to protect their genetic resources as they will serve as breeding material for developing new varieties in the future. Methods: Three breeds including Inner Mongolia cashmere goats (IMCG), Hanshan White cashmere goats (HS), and Ujimqin white cashmere goats (WZMQ) were studied. IMCG were of three types: Aerbas (AEBS), Erlangshan (ELS), and Alashan (ALS). Nine DNA samples were collected for each population, and they were genomically re-sequenced to obtain high-depth data. The genetic diversity parameters of each population were estimated to determine selection intensity. Principal component analysis, phylogenetic tree construction and genetic differentiation parameter estimation were performed to determine genetic relationships among populations. Results: Samples from the 45 individuals from the five goat populations were sequenced, and 30,601,671 raw single nucleotide polymorphisms (SNPs) obtained. Then, variant calling was conducted using the reference genome, and 17,214,526 SNPs were retained after quality control. Individual sequencing depth of individuals ranged from 21.13× to 46.18×, with an average of 28.5×. In the AEBS, locus polymorphism (79.28) and expected heterozygosity (0.2554) proportions were the lowest, and the homologous consistency ratio (0.1021) and average inbreeding coefficient (0.1348) were the highest, indicating that this population had strong selection intensity. Conversely, ALS and WZMQ selection intensity was relatively low. Genetic distance between HS and the other four populations was relatively high, and genetic exchange existed among the other four populations. Conclusion: The Inner Mongolia cashmere goat (AEBS type) population has a relatively high selection intensity and a low genetic diversity. The IMCG (ALS type) and WZMQ populations had relatively low selection intensity and high genetic diversity. The genetic distance between HS and the other four populations was relatively high, with a moderate degree of differentiation. Overall, these genetic variations provide a solid foundation for resource identification of Inner Mongolia Autonomous Region cashmere goats in the future.

• Asian-Australasian Journal of Animal Sciences
• /
• 제28권12호
• /
• pp.1686-1695
• /
• 2015

Molecular marker selection has been an acceptable tool in the acceleration of the genetic response of desired traits to improve production performance in chickens. The crossbreds from commercial parent stock (PS) broilers with four Thai synthetic breeds; Kaen Thong (KT), Khai Mook Esarn (KM), Soi Nin (SN), and Soi Pet (SP) were used to study the association among chicken growth hormones (cGH) and the insulin-like growth factor (IGF-I) genes for growth and carcass traits; for the purpose of developing a suitable terminal breeding program for Thai broilers. A total of 408 chickens of four Thai broiler lines were genotyped, using polymerase chain reaction-restriction fragment length polymorphism methods. The cGH gene was significantly associated with body weight at hatching; at 4, 6, 8, 10 weeks of age and with average daily gain (ADG); during 2 to 4, 4 to 6, 0 to 6, 0 to 8, and 0 to 10 weeks of age in $PS{\times}KM$ chickens. For $PS{\times}KT$ populations, cGH gene showed significant association with body weight at hatching, and ADG; during 8 to 10 weeks of age. The single nucleotide polymorphism variant confirmed that allele G has positive effects for body weight and ADG. Within carcass traits, cGH revealed a tentative association within the dressing percentage. For the IGF-I gene polymorphism, there were significant associations with body weight at hatching; at 2, 4, and 6 weeks of age and ADG; during 0 to 2, 4 to 6, and 0 to 6 weeks of age; in all of four Thai broiler populations. There were tentative associations of the IGF-I gene within the percentages of breast muscles and wings. Thus, cGH gene may be used as a candidate gene, to improve growth traits of Thai broilers.