• 대한음성학회지:말소리
• /
• 제56호
• /
• pp.207-223
• /
• 2005

The goal of language model adaptation is to improve the background language model with a relatively small adaptation corpus. This study presents a language model adaptation technique where additional text data for the adaptation do not exist. We propose the information retrieval (IR) technique with N-gram language modeling to collect the adaptation corpus from baseline text data. We also propose to use a dynamic language model interpolation coefficient to combine the background language model and the adapted language model. The interpolation coefficient is estimated from the word hypotheses obtained by segmenting the input speech data reserved for held-out validation data. This allows the final adapted model to improve the performance of the background model consistently The proposed approach reduces the word error rate by $13.6\%$ relative to baseline 4-gram for two-hour broadcast news speech recognition.

• Journal of applied mathematics & informatics
• /
• 제9권2호
• /
• pp.879-890
• /
• 2002

In this paper we consider open queueing system with a Poisson arrival process which have a finite upper bound on the arrival rate for which the system is stable. Interpolation approximations for quantities of interest, such as moments of the sojourn time distribution, have previously been developed for such systems, utilizing exact light and heavy traffic limits. These limits cannot always be easily computed for complex systems. Thus we consider an interpolation approximation where all of the relevant information is estimated via simulation. We show that all the relevant information can in fact be simultaneously estimated in a single regenerative simulation at any arrival rate. In addition to light and heavy traffic limits, both the quantity of interest and its derivative (with respect to the arrival rate) are estimated at the arrival rate of the simulation. All of the estimates are then combined, using a least squares procedure, to provide an interpolation approximation.

• Journal of Electrical Engineering and Technology
• /
• 제10권2호
• /
• pp.676-687
• /
• 2015

Path planning algorithms are used to allow mobile robots to avoid obstacles and find ways from a start point to a target point. The general path planning algorithm focused on constructing of collision free path. However, a high continuous path can make smooth and efficiently movements. To improve the continuity of the path, the searched waypoints are connected by the proposed polynomial interpolation. The existing polynomial interpolation methods connect two points. In this paper, point groups are created with three points. The point groups have each polynomial. Polynomials are made by matching the differential values and simple matrix calculation. Membership functions are used to distribute the weight of each polynomial at overlapped sections. As a result, the path has $G^2$ continuity. In addition, the proposed method can analyze path numerically to obtain curvature and heading angle. Moreover, it does not require complex calculation and databases to save the created path.

• 한국농공학회논문집
• /
• 제53권5호
• /
• pp.1-7
• /
• 2011

The objective of this study is to decide a spatial interpolation method on temperature data for the suitability analysis of garlic cultivation. In Korea, garlic is the second most cultivated condiment vegetable after red pepper. Nowadays warm-temperate garlic faces potential shift of its arable area according to warmer temperature in the Korean Peninsula, and the change can be drawn with the precise temperature map derived from interpolation on point-measured data. To find the preferable interpolation method in cases of germination and vegetative period of the garlic, different approaches were tested as follows: Inverse Distance Weighted (IDW), Spline, Ordinary Kriging (OK), and Universal Kriging (UK). As a result, IDW and UK show the lowest root mean square errors as for the germination and vegetative seasons, respectively. However, statistically significant difference was not revealed among the applied methods regarding the germinating period. Eventually this will contribute to mapping the suitable lands for the cultivation of warm-temperate garlic reasonably.

• 한국산학기술학회논문지
• /
• 제7권4호
• /
• pp.684-689
• /
• 2006

강우자료는 수문시스템 해석에 있어 가장 기본이 되는 입력자료이며, 강우측정시에는 다양한 원인에 의해 결측이 발생하게 된다 따라서, 이러한 자료를 보정하기 위한 다양한 방법들이 제시되어 있으나 적용성이나 오차정도에 대한 평가 없이 사용되고 있는 실정이다. 본 연구에서는 기존에 사용중인 산술평균법, 정상연강우량법, 수정정상연강우량법, 역거릭법, 선형계획법, 크리깅방법 등의 강우량 보정방법을 비교 평가하였다.

• Water Engineering Research
• /
• 제5권3호
• /
• pp.123-131
• /
• 2004

This paper describes the use of spatial interpolation for estimating seasonal crop potential evapotranspiration (PET) and irrigation water requirement in unmeasured evaporation gage stations within Edwards Aquifer, Texas using GIS. The Edwards Aquifer area has insufficient data with short observed records and rare gage stations, then, the investigation of data for determining of irrigation water requirement is difficult. This research shows that spatial interpolation techniques can be used for creating more accurate PET data in unmeasured region, because PET data are important parameter to estimate irrigation water requirement. Recently, many researchers are investigating intensively these techniques based upon mathematical and statistical theories. Especially, three techniques have well been used: Inverse Distance Weighting (IDW), spline, and kriging (simple, ordinary and universal). In conclusion, the result of this study (Table 1) shows the kriging interpolation technique is found to be the best method for prediction of unmeasured PET in Edwards aquifer, Texas.

• 한국CDE학회논문집
• /
• 제19권2호
• /
• pp.119-128
• /
• 2014

Computing the natural-looking interpolation of different shapes is a fundamental problem of computer graphics. It is proved by some researchers that such an interpolation can be achieved by pursuing the isometry. In this paper, a novel coordinate system that is invariant under isometries is defined. The coordinate system can easily be converted from the global vertex coordinates. Furthermore, the global coordinates can be efficiently recovered from the new coordinates by simply solving two sparse least-squares problems. Since the proposed coordinate system is invariant under isometries, then transformations such as global rigid trans-formations, articulated posture deformations, or any other isometric deformations, do not change the coordinate values. Therefore, shape interpolation can be done in this framework without being affected by the distortions caused by the isometry.

• Structural Engineering and Mechanics
• /
• 제59권6호
• /
• pp.1071-1094
• /
• 2016

We employ the so-called problem-dependent linked interpolation concept to develop two cubic 4-node quadrilateral plate finite elements with 12 external degrees of freedom that pass the constant bending patch test for arbitrary node positions of which the second element has five additional internal degrees of freedom to get polynomial completeness of the cubic form. The new elements are compared to the existing linked-interpolation quadratic and nine-node cubic elements presented by the author earlier and to the other elements from literature that use the cubic linked interpolation by testing them on several benchmark examples.

• 한국정보디스플레이학회:학술대회논문집
• /
• 한국정보디스플레이학회 2006년도 6th International Meeting on Information Display
• /
• pp.1259-1262
• /
• 2006

In this paper, we propose a modified cubic convolution interpolation for the enlargement or reduction of digital images using a pixel difference value. The proposed method has a low complexity: the number of multiplier of weighted value to calculate one pixel of a scaled image has seven less than that of cubic convolution interpolation has sixteen. We use the linear function of the cubic convolution and the difference pixel value for selecting interpolation methods. The proposed method is compared with the conventional one for the computational complexity and the image quality. The simulation results show that the proposed method has less computational complexity than one of the cubic convolution interpolation.

• Journal of Mechanical Science and Technology
• /
• 제15권2호
• /
• pp.192-198
• /
• 2001

We consider an interpolation problem with nonuniform rational B-spline curves given ordered data points. The existing approaches assume that weight for each point is available. But, it is not the case in practical applications. Schneider suggested a method which interpolates data points by automatically determining the weight of each control point. However, a drawback of Schneiders approach is that there is no guarantee of avoiding undesired poles; avoiding negative weights. Based on a quadratic programming technique, we use the weights of the control points for interpolating additional data. The weights are restricted to appropriate intervals; this guarantees the regularity of the interpolating curve. In a addition, a knot placement is proposed for pleasing interpolation. In comparison with integral B-spline interpolation, the proposed scheme leads to B-spline curves with fewer control points.