• Animal Bioscience
• /
• v.37 no.4
• /
• pp.622-630
• /
• 2024

Objective: Pig breeders cannot obtain phenotypic information at the time of selection for sow lifetime productivity (SLP). They would benefit from obtaining genetic information of candidate sows. Genomic data interpreted using deep learning (DL) techniques could contribute to the genetic improvement of SLP to maximize farm profitability because DL models capture nonlinear genetic effects such as dominance and epistasis more efficiently than conventional genomic prediction methods based on linear models. This study aimed to investigate the usefulness of DL for the genomic prediction of two SLP-related traits; lifetime number of litters (LNL) and lifetime pig production (LPP). Methods: Two bivariate DL models, convolutional neural network (CNN) and local convolutional neural network (LCNN), were compared with conventional bivariate linear models (i.e., genomic best linear unbiased prediction, Bayesian ridge regression, Bayes A, and Bayes B). Phenotype and pedigree data were collected from 40,011 sows that had husbandry records. Among these, 3,652 pigs were genotyped using the PorcineSNP60K BeadChip. Results: The best predictive correlation for LNL was obtained with CNN (0.28), followed by LCNN (0.26) and conventional linear models (approximately 0.21). For LPP, the best predictive correlation was also obtained with CNN (0.29), followed by LCNN (0.27) and conventional linear models (approximately 0.25). A similar trend was observed with the mean squared error of prediction for the SLP traits. Conclusion: This study provides an example of a CNN that can outperform against the linear model-based genomic prediction approaches when the nonlinear interaction components are important because LNL and LPP exhibited strong epistatic interaction components. Additionally, our results suggest that applying bivariate DL models could also contribute to the prediction accuracy by utilizing the genetic correlation between LNL and LPP.

• Journal of Intelligence and Information Systems
• /
• v.1 no.2
• /
• pp.1-16
• /
• 1995

The field of human judgment and decision making provides useful methodologies for examining the human decision making process and substantive results. One of the methodologies is a lens model analysis which can examine valid nonlinearity in the human decision making process. Using the method, valid nonlinearity in human decision behavior can be successfully detected. Two linear(statistical) models of human experts and two nonlinear models of human experts are compared in terms of predictive accuracy (predictive validity). The results indicate that nonlinear models can capture factors(valid nonlinearity) that contribute to the expert's predictive accuracy, but not factors (inconsistency) that detract from their predictive accuracy. Then, it is argued that nonlinear models cab be more accurate than linear models, or as accurate as human experts, especially when human experts employ valid nonlinear strategies in decision making.

• Korean Journal of Applied Biomechanics
• /
• v.28 no.2
• /
• pp.127-134
• /
• 2018

Objective: In a statistical linear model estimating the center of rotation of a human hip joint, which is the parameter related to the mean of response vectors, assumptions of homoscedasticity and independence of position vectors measured repeatedly over time in the model result in an inefficient parameter. We, therefore, should take into account the variance-covariance structure of longitudinal responses. The purpose of this study was to estimate the efficient center of rotation vector of the hip joint by using covariance pattern models. Method: The covariance pattern models are used to model various kinds of covariance matrices of error vectors to take into account longitudinal data. The data acquired from functional motions to estimate hip joint center were applied to the models. Results: The results showed that the data were better fitted using various covariance pattern models than the general linear model assuming homoscedasticity and independence. Conclusion: The estimated joint centers of the covariance pattern models showed slight differences from those of the general linear model. The estimated standard errors of the joint center for covariance pattern models showed a large difference with those of the general linear model.

• Asian-Australasian Journal of Animal Sciences
• /
• v.9 no.4
• /
• pp.439-447
• /
• 1996

Monthly test day production for 12,020 records, were collected from six of the largest specialized dairy farms located in central region of the Kingdom of Saudi Arabia. The records described lactating cows in four parities and two seasons of calving. Monthly test day records were fitted using Wood's model $At{{^b}{_e}}^{-ct}$ with multiple and additive error term. Linear and non-linear regression models were used to find the estimates of the parameters necessary to draw the lactation curves. The shape of the lactation curves of different parities showed that third lactation has the heighest peak (43.08 kg) for linear regression model and (42.08 kg) for non-linear regression model. Fourth lactation has the lowest peak (24.00kg) for linear regression model and (25.64 kg) for non-linear regression models. Cows of second and third lactations reached the peak at 58 day for both linear and non-linear regression models. Cows of first lactation were more persistent and had late peak at 68 and 67 days for both models respectively. While, third lactation cows were lower persistent and had early peak at 58 day for both models. Cows calved at winter months have higher starting values (A), higher ascending slope (b) and higher decending slope (c). Least square means of milk yield of the first four parities and for overall data were 6,653, 7,659, 7,482, 6,988 and 7,614 kg respectively. The corresponding lactation period were 358, 367, 350, 363 and 364 days respectively.

• Journal of Korean Society of Water and Wastewater
• /
• v.21 no.5
• /
• pp.601-607
• /
• 2007

In order to maintain constant residual chlorine in sedimentation basin, It is necessary to develop real time prediction model of residual chlorine considering water treatment plant data such as water qualities, weather, and plant operation conditions. Based on the operation data acquired from K water treatment plant, prediction models of residual chlorine in sediment basin were accomplished. The input parameters applied in the models were water temperature, turbidity, pH, conductivity, flow rate, alkalinity and pre-chlorination dosage. The multiple regression models were established with linear and non-linear model with 5,448 data set. The corelation coefficient (R) for the linear and non-linear model were 0.39 and 0.374, respectively. It shows low correlation coefficient, that is, these multiple regression models can not represent the residual chlorine with the input parameters which varies independently with time changes related to weather condition. Artificial neural network models are applied with three different conditions. Input parameters are consisted of water quality data observed in water treatment process based on the structure of auto-regressive model type, considering a time lag. The artificial neural network models have better ability to predict residual chlorine at sediment basin than conventional linear and nonlinear multi-regression models. The determination coefficients of each model in verification process were shown as 0.742, 0.754, and 0.869, respectively. Consequently, comparing the results of each model, neural network can simulate the residual chlorine in sedimentation basin better than mathematical regression models in terms of prediction performance. This results are expected to contribute into automation control of water treatment processes.

• Proceedings of the Earthquake Engineering Society of Korea Conference
• /
• 2006.03a
• /
• pp.519-526
• /
• 2006

This paper compares the seismic response control performance of linear and non-linear models fer tuned liquid damper (TLD). The existing linear and nonlinear TLD models were used for the numerical analysis of single degree of freedom (SDOF) and multi degree of freedom (MDOF) systems with TLD. The nonlinear model considers the variation of the frequency and damping of the TLD with varying excitation amplitude while the linear one has the invariant parameters. Numerical analysis results from SDOF systems indicate that the nonlinear model shows about 5% better control performance than linear one when the mass ratio is 2% and the optimal parameters for reducing peak responses are dependent on the characteristics of the excitation earthquake loads.

• Communications for Statistical Applications and Methods
• /
• v.22 no.5
• /
• pp.507-518
• /
• 2015

It is not easy to select a linear mixed model since the main interest for model building could be different and the number of parameters in the model could not be clearly defined. In this paper, performance of conditional Akaike Information Criteria and its bias-corrected version are compared with marginal Bayesian and Akaike Information Criteria through a simulation study. The results from the simulation study indicate that bias-corrected conditional Akaike Information Criteria shows promising performance when candidate models exclude large models containing the true model, but bias-corrected one prefers over-parametrized models more intensively when a set of candidate models increases. Marginal Bayesian and Akaike Information Criteria also have some difficulty to select the true model when the design for random effects is nested.

• Communications for Statistical Applications and Methods
• /
• v.14 no.1
• /
• pp.229-239
• /
• 2007

We present a diagnostic method for the quasi-likelihood estimators in generalized linear models. Since these estimators can be usually obtained by iteratively reweighted least squares which are well known to be very sensitive to unusual data, a diagnostic step is indispensable to analysis of data. We extend the local influence approach based on the maximum likelihood function to that on the quasi-likelihood function. Under several perturbation schemes local influence diagnostics are derived. An illustrative example is given and we compare the results provided by local influence and deletion.

• Journal of the Korean Data and Information Science Society
• /
• v.9 no.2
• /
• pp.219-225
• /
• 1998

The parameters in linear models with censored normal responses are usually estimated by the iterative maximum likelihood and least square methods. However, the iterative least square method is simple but hardly has theoretical justification, and the iterative maximum likelihood estimating equations are complicatedly derived. In this paper, we justify these methods via Wedderburn (1974)'s quasi-likelihood approach. This provides an explicit justification for the iterative least square method and also directly the iterative maximum likelihood method for estimating the regression coefficients.

• 제어로봇시스템학회:학술대회논문집
• /
• 1996.10b
• /
• pp.241-244
• /
• 1996

This paper discusses the general properties and the design procedures of Internal Model Control(IMC) scheme for nonlinear plants. Also we propose new nonlinear IMC(NIMC) design method using linear IMC. Although all IMC controllers can be thought simple 'inverse controller', its nonlinear realization is not easy. Propose NIMC is composed multiple linear models, IMC controllers, and switching scheme. The advantages of this method are we can use simple linear IMC design method and need not nonlinear modelings.