• MALSORI
• /
• no.56
• /
• 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.

• The Journal of Engineering Geology
• /
• v.18 no.1
• /
• pp.27-36
• /
• 2008

To derive easily the coefficient of permeability from several other soil properties, the estimation model of coefficient of permeability was proposed using linear regression analysis. The coefficient of permeability is one of the major factors to evaluate the soil characteristics. The study area is located in Kangwon-do Pyeongchang-gun Jinbu-Myeon. Soil samples of 45 spots were taken from the study area and various soil tests were carried out in laboratory. After selecting the soil factor influenced by the coefficient of permeability through the correlation analysis, the estimation model of coefficient of permeability was developed using the linear regression analysis between the selected soil factor and the coefficient of permeability from permeability test. Also, the estimation model of coefficient of permeability was compared with the results from permeability test and empirical equation, and the suitability of proposed model was proved. As the result of correlation analysis between various soil factors and the coefficient of permeability using SPSS(statistical package for the social sciences), the largest influence factor of coefficient of permeability were the effective grain size, porosity and dry unit weight. The coefficient of permeability calculated from the proposed model was similar to that resulted from permeability test. Therefore, the proposed model can be used in case of estimating the coefficient of permeability at the same soil condition like study area.

• Tribology and Lubricants
• /
• v.19 no.4
• /
• pp.230-236
• /
• 2003

In applications such as MEMS and NEMS devices, the adhesion force and contact load may be of the same order of magnitude and the static friction coefficient can be very large. Such large coefficient may result in unacceptable and possibly catastrophic adhesion, stiction, friction and wear. To obtain the static friction coefficient of contacting real surfaces without the assumption of an empirical coefficient value, numerical simulations of the contact load, tangential force, and adhesion force are preformed. The surfaces in dry contact are statistically modeled by a collection of spherical asperities with Gaussian height distribution. The asperity micro-contact model utilized in calculation (the ZMC model), considers the transition from elastic deformation to fully plastic flow of the contacting asperity. The force approach of the modified DMT model using the Lennard-Jones attractive potential is applied to characterize the intermolecular forces. The effect of the surface topography on the static friction coefficient is investigated for cases rough, intermediate, smooth, and very smooth, respectively. Results of the static friction coefficient versus the external force are presented for a wide range of plasticity index and surface energy, respectively. Compared with those obtained by the GW and CEB models, the ZMC model is more complete in calculating the static friction coefficient of rough surfaces.

• Proceedings of the Korean Society for Technology of Plasticity Conference
• /
• 1999.08a
• /
• pp.308-312
• /
• 1999

Developed the model equations which calculate roll force, roll power during hot rolling in real time. The variables which mainly effect on the roll force, roll power are shape factor, reduction, roll diameter, roll velocity, strip inlet temperature, carbon content of strip and strip-roll contact friction coefficient. Among these variables roll diameter, roll velocity, inlet temperature, carbon content and friction coefficient can be excluded in interpolated model equation by introducing equation of die force(F'), power(p') of the frictionless uniform plane strain compression which can be calculated without iteration. At the case of coulomb friction coefficient of 0.3, we evaluated coefficient of polynomial equations of {{{{ { F} over {F' } }}}}, {{{{ { Pf} over {Pd }, { Pd} over {P' } }}}} from the result of finite element analysis using interpolation. It was found that the change of values of {{{{ { F} over {F' }, { P} over {P' } }}}} with the friction coefficient tend to straight line which slope depend only on shape factor. With these properties, developed model equations could be extended to other values of coulomb friction coefficient. To verify developed roll force, roll power model equation we compared the results from these model equation with the results from these model equation with the results from finite element analysis in factory process condition.

• Journal of the Korea Safety Management & Science
• /
• v.19 no.1
• /
• pp.157-167
• /
• 2017

The two-way quality theory has been widely used as a method for classifying quality attributes for several decades. In particular, the Kano model that classifies attributes into not just conventional one-dimensional but must-be and attractive has gained popularity due to its applicability and ease of use. However, the wordings of the five alternatives in the Kano's questionnaire has been criticised for unclear meanings. This study proposes a new two-way model to classify attributes using 5-point Likert scale alternatives. For this, the current paper investigated a case of TV sets to examine how the proposed model works in comparison with the Kano model. The application results of the proposed model are different from the original one. The two-way model classifies quality attributes in more detail such as the "one-dimensional with an attractive tendency" attribute, which has a greater influence on satisfaction than dissatisfaction, the opposite "one-dimensional with a must-be tendency" attribute, and "highly one-dimensional" and "less one-dimensional" attributes. In this study, a potential satisfaction coefficient (PSC), a potential dissatisfaction coefficient (PDC), and an average potential coefficient (APC) to manage quality attributes are proposed and discussed for their utilization.

• Coupled systems mechanics
• /
• v.11 no.1
• /
• pp.1-14
• /
• 2022

The key parameter that affects the consolidation process of soil is the coefficient of permeability. The common assumption in the consolidation analysis is that the coefficient of permeability is porosity-dependent. However, various authors suggest that the strain-dependency of the coefficient of permeability should also be taken into account. In this paper, we present results of experimental and numerical analyses, with an aim to determine the strain-dependency of the coefficient of permeability. We present in detail both the experimental procedure and the finite element formulation of the two-dimensional axisymmetric numerical model of the oedometer test (standard and modified). We perform a set of experimental standard and modified oedometer tests. We use these experimental results to validate our numerical model and to define the model input parameter. Finally, by combining the experimental and numerical results, we propose the expression for the strain-dependent coefficient of permeability.

• Journal of the Korean Data and Information Science Society
• /
• v.20 no.5
• /
• pp.787-796
• /
• 2009

A random coefficient model is applied when times of the repeated measurements are not fixed in experiments with respect to the subjects. The procedures of the inference of a random coefficient model are same as those of a mixed model. Diet experimental data was used for applying the random coefficient model. Various random coefficient models are investigated for the experimental data, and are compared each other. Finally, optimal random coefficient model would be selected. It resulted from the analysis that for the fixed effect factor, the baseline, treatment, height, and time effect were very significant. The treatment effect of the diet foods and exercises were more effective in losing weight than the effect of the diet foods only. The fixed cubic time effect was very significant. The variance components corresponding to the subject effect, linear time effect, quadratic time effect, and cubic time effect of the random coefficients are all positive. When quartic time effect was added as random coefficients the model did not converge. Thus random coefficients up to the cubic terms was considered as the optimal model.

• Journal of ILASS-Korea
• /
• v.20 no.1
• /
• pp.53-58
• /
• 2015

In this study, we prepare the Ag nanofluids synthesized by the chemical reduction method and measure the extinction coefficient of those nanofluids at a wavelength of 632.8 nm. The Ag nanofluids are synthesized by the chemical reduction method using silver nitrate ($AgNO_3$) and sodium borohydride ($NaBH_4$) in water and ethylene glycol (EG). For stable dispersion of Ag particles in the base liquids, polyvinyl pyrrolidone (PVP) is added as a surfactant. The extinction coefficient of manufactured Ag nanofluids is measured by an in-house developed measurement system at the wavelength of 632.8 nm. The results show that the extinction coefficient of water-based and EG-based Ag nanofluids is linearly increased with respect to the particle loadings. Moreover, it is shown that the extinction coefficient of EG-based Ag nanofludis is higher than that of water-based Ag nanofluids. Finally we compare the experimental results with both the Maxwell-Garnett model and Rayleigh scattering approximation model, and they demonstrate that the Rayleigh scattering approximation model is reasonably predict the extinction coefficient of Ag nanofluids using hydraulic diameter of silver nanoparticle.

• Journal of the Korean Data and Information Science Society
• /
• v.27 no.4
• /
• pp.1059-1066
• /
• 2016

In this paper we propose a robust version of varying coefficient models, which is based on the regularized regression with L1 regularization. We use the iteratively reweighted least squares procedure to solve L1 regularized objective function of varying coefficient model in locally weighted regression form. It provides the efficient computation of coefficient function estimates and the variable selection for given value of smoothing variable. We present the generalized cross validation function and Akaike information type criterion for the model selection. Applications of the proposed model are illustrated through the artificial examples and the real example of predicting the effect of the input variables and the smoothing variable on the output.

• Journal of the Korean Institute of Illuminating and Electrical Installation Engineers
• /
• v.28 no.11
• /
• pp.51-57
• /
• 2014

This paper presents an algorithmic type computing technique of process coefficient in predicting model of temperature for reheating furnace and also suggests a design method of neural network model to find an adequate value of process coefficient for arbitrary operating conditions including test conditons. The proposed neural network use furnace temperature, line speed and slab information as input variables, and process coefficient is output variable. Reasonable process coefficients can be obtained by an algorithmic procedure proposed in this paper using process data gathered at test conditons. Also, neural network model output equal process coefficient under same input conditions. This means that adquate process coefficients can be found by only computing neural network model without additive test even if operating conditions vary.