• Journal of KIISE:Software and Applications
• /
• v.26 no.12
• /
• pp.1383-1392
• /
• 1999

이미 학습된 다층퍼셉트론 신경망을 디지털 VLSI 기술을 사용하여 하드웨어로 구현할 경우 신경망의 가중치 및 뉴런 출력들을 양자화해야 하는 문제가 발생한다. 이러한 신경망 변수들의 양자화는 결과적으로 주어진 입력에 대한 신경망의 최종 출력에서의 왜곡을 초래한다. 본 논문에서는 먼저 이러한 양자화로 인한 신경망 출력에서의 왜곡을 통계적으로 분석하였다. 분석 결과에 의하면 입력패턴 각 성분의 제곱들의 합과 가중치의 크기들이 양자화 영향에 주로 기여하는 것으로 나타났다. 이러한 분석 결과를 이용하여 양자화를 위한 정밀도가 주어졌을 때, 양자화 영향이 최소화된 다층퍼셉트론 신경망을 설계하는 방법을 제시하였다. 그리고 제안된 방법에 의해 얻은 신경망과 오류역전파 학습방법에 의하여 얻은 신경망의 성능을 비교함으로써 제안된 방법의 효율성을 입증하였다. 실험결과는 낮은 양자화 정밀도에서도 제안된 방법이 더 좋은 성능을 보였다.Abstract When we implement a multilayer perceptron with the digital VLSI technology, we generally have to quantize the weights and the neuron outputs. These quantizations eventually cause distortion in the output of the network for a given input. In this paper first we made a statistical analysis about the effect caused by the quantization on the output of the network. The analysis revealed that the sum of the squared input components and the sizes of the weights are the major factors which contribute to the quantization effect. We present a design method for an MLP which minimizes the quantization effect when the precision of the quantization is given. In order to show the effectiveness of the proposed method, we developed a network by our method and compared it with the one developed by the regular backpropagation. We could confirm that the network developed by our method performs better even with a low precision of the quantization.

• The Korean Journal of Applied Statistics
• /
• v.31 no.5
• /
• pp.587-597
• /
• 2018

Procedures, such as sampling technique, survey method, and questionnaire preparation, are required in order to obtain sample data in accordance with the purpose of a survey. An important procedure is the decision of the sample size formula. The sample size formula is determined by setting the target error and total cost according to the sampling method. In this paper, we propose a sample size formula using population changes over time, estimation error of the previous time and response rate of past data when the target error and the expected response rate are given in the simple random sampling. In actual research, we use estimators that apply complex weights in addition to design-based weights. Therefore, we induce a sample size formula for estimators using design-based weights and nonresponse adjustment coefficients, that can be a formula that reflects differences in response rates when survey methods are changed over time. In addition, we use simulations to compare the proposed formula with the existing sample size formula.

• Korean Journal of Human Ecology
• /
• v.13 no.1
• /
• pp.153-158
• /
• 2004

With considerable development of comfortable and functional clothing in recent years, we need to evaluate the effects of garment pressure in daily wear on each parts of human body because the garment pressure is important to design the clothing. This study was designed to examine the effects of body postures on garment pressure on each parts of human body in the actual clothing conditions. All the data were collected from 50 volunteered subjects. The Garment pressure was measured in lune and December with 8 points CPMS clothing pressure system from scapular, upper am, elbow, under arm, front waist line, side waist line, abdomen, crista ilica, upper hip, middle hip, front thigh, back thigh, front knee and back knee. The postures of subjects were controlled with 3 positions such as standing (posture 1), sitting on the chair (posture 2), and sitting on the floor (posture 3) during measurement of clothing pressure. Clothing weights were more in men than in woman. It showed that clothing weights had no effects on the garment pressure. In this study, however, just the garment pressures on scapular and top of the hip increased significantly by clothing weight (p<. 05). Clothing horizontally pressed on scapular and top of hip but not on other parts. When subjects stood up, the garment pressure was the highest on the side waist. Especially, clothing pressure on the front waist point was lower than that of the left side waist. On the upper parts of the human body, the garment pressure of left side waist was the highest, and followed by front waist, crista ilica, and abdomen in order. When subjects were sitting on the chair, the garment pressure on the lower parts of the human body was the highest on the top of hip. When the subjects were sitting on the chair or on the floor, the surface area on their skin of hip and waist parts increased by postures. In addition, it showed that men felt more comfortable than women on higher clothing pressure level.

• Proceedings of the KIEE Conference
• /
• 1999.07b
• /
• pp.936-938
• /
• 1999

In this Paper, the design of a feedback linearization controller using multilayer neural network is proposed. The Proposed feedback linearization control scheme is designed by finding Lie derivatives from an identified neural networks. Lie derivatives are expressed as a combination of weights and neuron outputs. The proposed method is applied to an antenna arm problem and the simulation results show performance comparisons between the ordinary feedback linearization and the Proposed method.

• 제어로봇시스템학회:학술대회논문집
• /
• 1996.10a
• /
• pp.334-337
• /
• 1996

This is a paper intended for initial stage of vehicle modeling and design. The needs to determine a variety of vehicle suspension parameters required for initial design has been difficult and time-consuming task. In order to facilitate a concise and efficient presentation of initial vehicle design procedure, this paper uses a mathematical model and physical geometry. Vehicle model consists of dimensions, inertias and mechanical constants. These vehicle model parameters divided into several categories : basic parameters, coefficients and constants, design specification, spring and damper, bush stiffness, stabilizer bar, suspension geometry, tire, and vehicle weights of various design condition. This paper uses a vehicle design fundamental (VDF) program running under Windows 95 graphical interface. The features of VDF will be briefly outlined in this paper.

• Asian-Australasian Journal of Animal Sciences
• /
• v.22 no.5
• /
• pp.686-693
• /
• 2009

A 21- to 42-day feeding study was conducted in Ross male broilers to evaluate the use of dried bakery product (DBP) and the influence of adding fat at different points in the manufacturing process. Six dietary treatments were formulated using a factorial arrangement (3${\times}$2 design) with three levels of fat in the mixer (high: 4.8%, medium: 3.8% and low: 2.8%) with or without DBP (0 and 7%). Additional fat was sprayed on pellets in a post-pelleting liquid application to bring the fat content to a similar level in all diets. Data on pellet quality (before and after post-pelleting fat addition), broiler performance, nutrient digestibility and organ weights were studied. Pellets made with DBP showed higher hardness values when measured before post-pelleting fat addition (p<0.001), although DBP did not affect final pellet hardness or durability. Higher post-pelleting hardness and durability were shown by diets to which a lower level of fat had been added in the mixer (p<0.001). In general, post-pelleting fat application improved durability (p<0.05). However, broiler performance and ileal digestibility were not affected by any of the factors tested. Dietary treatments had a significant but variable effect on carcass yield (p<0.01), although there were no differences among treatments regarding breast and leg yield, abdominal fat or organ weights. The results indicate that up to 7% DBP could be used in the broiler diet without impairing performance, ileal digestibility or organ weights. The place or point of fat addition in the manufacturing process has a strong influence on pellet quality.

• Proceedings of the KSME Conference
• /
• 2004.11a
• /
• pp.840-845
• /
• 2004

The reliability analysis for nonnormal distributions using the three level DOE(design of experiments) method was developed by Seo and Kwak in 2002. Although this method estimates only up to the first four moments(mean, standard deviation, skewness, and kurtosis) of the system response function, the result and the type of probability distribution determined by using the Pearson system are shown very good. However the accuracy is low in case of nonlinear performance function and sometimes, the level calculated is outside of the region in which the random variable is defined. In this article we suggest a modified three level DOE method to overcome these weaknesses and to obtain optimum choice for 3 levels and weights to handle nonnormal distributions. Furthermore we extend it to finding the optimum choice for 5 levels and weights to increase the accuracy in case of nonlinear performance function. A systematic procedure for reliability analysis is then proposed by using the Pearson system.

• Journal of the Korean Operations Research and Management Science Society
• /
• v.32 no.1
• /
• pp.61-75
• /
• 2007

The purpose of this paper is to develop a new group classification method based on genetic algorithm and to com-pare its prediction performance with those of existing methods in the area of bond rating. To serve this purpose, we conduct various experiments with pilot and general models. Specifically, we first conduct experiments employing two pilot models : the one searching for the cluster center of each group and the other one searching for both the cluster center and the attribute weights in order to maximize classification accuracy. The results from the pilot experiments show that the performance of the latter in terms of classification accuracy ratio is higher than that of the former which provides the rationale of searching for both the cluster center of each group and the attribute weights to improve classification accuracy. With this lesson in mind, we design two generalized models employing genetic algorithm : the one is to maximize the classification accuracy and the other one is to minimize the total misclassification cost. We compare the performance of these two models with those of existing statistical and artificial intelligent models such as MDA, ANN, and Decision Tree, and conclude that the genetic algorithm-based group classification method that we propose in this paper significantly outperforms the other methods in respect of classification accuracy ratio as well as misclassification cost.

• The Transactions of the Korean Institute of Electrical Engineers A
• /
• v.51 no.4
• /
• pp.188-195
• /
• 2002

We proposed the design of SVC adaptive fuzzy logic controller(AFLC) using Tabu search and neural network. We tuned the gains of input-output variables of fuzzy logic controller(FLC) and weights of neural network using Tabu search. Neural network was used for adaptively tuning the output gain of FLC. The weights of neural network was learned from the back propagation algorithm in real-time. To evaluate the usefulness of AFLC, we applied the proposed method to single-machine infinite system. AFLC showed the better control performance than PD controller and GAFLS[10] for three-phase fault in nominal load which had used when tuning AFLC. To show the robustness of AFLC, we applied the proposed method to disturbances such as three-phase fault in heavy and light load. AFLC showed the better robustness than PD controller and GAFLC[10].

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.65 no.6
• /
• pp.1070-1079
• /
• 2016

In this paper, the design of recursive radial basis function neural networks based on incremental fuzzy c-means is introduced for processing the big data. Radial basis function neural networks consist of condition, conclusion and inference phase. Gaussian function is generally used as the activation function of the condition phase, but in this study, incremental fuzzy clustering is considered for the activation function of radial basis function neural networks, which could effectively do big data processing. In the conclusion phase, the connection weights of networks are given as the linear function. And then the connection weights are calculated by recursive least square estimation. In the inference phase, a final output is obtained by fuzzy inference method. Machine Learning datasets are employed to demonstrate the superiority of the proposed classifier, and their results are described from the viewpoint of the algorithm complexity and performance index.