• Proceedings of the Korean Statistical Society Conference
• /
• 2000.11a
• /
• pp.219-225
• /
• 2000

데이터 마이닝 분야에서 널리 사용되고 있는 신경망은 최근 많은 통계인들의 관심을 끌고 있다. 그러나 범용 근사성(universal approximator)이라는 성질에도 불구하고 초기치에 따라 적합 결과가 크게 좌우되는 단점이 있다. 본 논문에서는 붓스트랩 표본을 통해 초기치를 발견하는 bumping 기법이 신경망 분야에서 사용되고 있는 무작위 탐색법 보다 더 정확하고 안정적인 초기치를 제공하여 주는가를 살펴 보았다.

• The Transactions of the Korea Information Processing Society
• /
• v.4 no.3
• /
• pp.747-753
• /
• 1997

Function approximation from a set of input-output parirs has numerous applications in scientiffc and engineer-ing areas.Multiayer feedforward neural networks have been proposed as a good approximator of noninear function.The back propagation (BP) algorithm allows muktiayer feedforward neural networks oro learn input-output mappongs from training samples.However, the mapping acquired through the BP algorithm nay be cor-rupt when errorneous trauning data are employed.In this paper we propose a robust BP learning algorithm that is resistant to the errormeous data and is capable of rejecting gross errors during the approximation process.

• Proceedings of the Korean Statistical Society Conference
• /
• 2003.05a
• /
• pp.19-24
• /
• 2003

신경망은 적용 다양성과 제약조건의 최소성, 강력한 예측성, 범용성, 근사성 등 많은 장점을 지니고 있으나 초기 가중치의 할당에 따라 모델 생성의 Performance와 예측의 결과가 달라지게 되는 단점을 지니고 있다. 이런 신경망의 초기 가중치에 따른 단점을 보안하기 위해 통계적 알고리즘의 접목을 통해 Hybrid된 신경망 보완 알고리즘을 제시하고자 하였다. 논문을 위한 기본 가정으로 신경망의 가장 기본인 SLP 알고리즘을 바탕으로 활성함수에 가장 일반적으로 사용되는 Sigmoid 활성함수를 이용하였을 때, 초기 가중치로 기존의 임의 난수 생성 방식이 아닌 통계적 로지스틱 회귀분석의 계수값(mle)을 제시하여 이를 초기치로 사용한 경우와 그렇지 않은 경우의 예측 정확성과 수렴의 Performance정도를 비교하여 가장 효과적인 초기치 방법을 제시하고자 하였다.

• Journal of KSNVE
• /
• v.9 no.4
• /
• pp.849-855
• /
• 1999

Neural Networks have shown potential to become an attractive alternative to classic methods for identification and control of non-linear dynamic systems. The purpose of this paper is to present an application of neural networks, that is a neural reconstruction of the input signal of a non-linear unknown system. This basic methodology could be used for practical purpose in several engineering fields. Clearly applications of the proposed scheme can be of interest for physical systems where a complete network of sensors measuring system inputs is not available. It should also be emphasized that the application of the reconstruction scheme is of little or no interest when the analyzed system works and operates at nominal conditions. In fact, only when failures and/or system anomailes occur, leasing to performance degradation and/or shutdown, the application of this scheme is of interest. The paper presents the results of the methodology applied to unknown non-linear dynamic systems and the robustness of the scheme to white and colored system noise was evaluated.

• Journal of The Korean Association of Information Education
• /
• v.3 no.1
• /
• pp.134-142
• /
• 1999

The demand for recognizing the color as well as the object shape is increasing to use the detailed information, because-the expense of color input/output devices become cheap. The research on the color correction should be researched for the exact color presentation and color reproduction of color input/output systems. In this paper, we researched on the color correction of color scanner. The characterization of color scanner is a two step process of gray-balancing and color transformation. The decoupling of the gray-balancing from the color transformation enables the portability of the scanner characterization. We used the least square methods for the line fitting and the Neural Network for the storage space and computation speed. The output of Neural Network is similar to the target value in three-dimensional tristimulus space. The proposed color correction method can be used for all scanners of a manufacturer's model because of the portability.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.10 no.1
• /
• pp.206-211
• /
• 2006

The quantification analysis problem is that how the m entities that have n characteristics can be linked to p-dimension space to reflect the similarity of each entity In this paper, the optimization approach for the quantification analysis problem using neural networks is suggested, and the performance is analyzed The computation of average variation volume by mean field theory that is analytical approximated mobility of a molecule system and the annealed mean field neural network approach are applied in this paper for solving the quantification analysis problem. As a result, the suggested approach by a mean field annealing neural network can obtain more optimal solution than the eigen value analysis approach in processing costs.

• Proceedings of the Korea Multimedia Society Conference
• /
• 2003.11b
• /
• pp.569-572
• /
• 2003

본 연구에서는 뉴우턴법에 기초한 고정점 알고리즘의 신경망 기반 독립성분분석에 kurtosis를 추가한 독립성분의 군집화를 제안하였다. 여기서 뉴우턴법의 고정점 알고리즘은 엔트로피에 기초한 목적 함수의 근을 구하는 근사화 방법으로 빠른 성분분석을 위함이고, kurtosis는 독립성분의 추출순서를 고려하지 않는 속성을 개선하기 위함이다. 제안된 기법을 256$\times$256 픽셀의 8개 혼합영상의 분리에 적용한 결과, 제안된 방법은 기존의 독립성분분석에서 분석순서를 고려치 않는 제약을 효과적으로 해결 할 수 있음을 확인하였다.

• Journal of the Korean Institute of Illuminating and Electrical Installation Engineers
• /
• v.13 no.2
• /
• pp.63-70
• /
• 1999

This paper is proposed a new method to deal with the optimized auto-tuning for the Pill controller which is used to the process-control in various fields. First of all, in this method, 1st order delay system with dead time which is modelled from the unit step response of the system is Pade-approximated, then initial values are determined by the Ziegler-Nichols method. So we can find the parameters of Pill controller so as to minimize the fuzzy criterion function which includes the maximum overshoot, damping ratio, rising time and settling time. Finally, after studying the parameters of Pill controller by Backpropagation of Neural-Network, when we give new K, L, T values to Neural-Network, the optimized parameter of Pill controller is found by Neural-Network Program.rogram.

• Journal of the Korean Data and Information Science Society
• /
• v.8 no.2
• /
• pp.163-171
• /
• 1997

Function approximation from a set of input-output pairs has numerous applications in scientific and engineering areas. Multilayer feedforward neural networks have been proposed as a good approximator of nonlinear function. The back propagation(BP) algorithm allows multilayer feedforward neural networks to learn input-output mappings from training samples. It iteratively adjusts the network parameters(weights) to minimize the sum of squared approximation errors using a gradient descent technique. However, the mapping acquired through the BP algorithm may be corrupt when errorneous training data we employed. When errorneous traning data are employed, the learned mapping can oscillate badly between data points. In this paper we propose a robust BP learning algorithm that is resistant to the errorneous data and is capable of rejecting gross errors during the approximation process, that is stable under small noise perturbation and robust against gross errors.

• Proceedings of the KIEE Conference
• /
• 2008.10b
• /
• pp.187-192
• /
• 2008

A saturation and deadzone compensator is designed for robot systems using fuzzy logic (FL) and neural network (NN). The classification property of FL system and the function approximation ability of the NN make them the natural candidate for the rejection of errors induced by the saturation and deadzone. The tuning algorithms are given for the fuzzy logic parameters and the NN weights, so that the saturation and deadzone compensation scheme becomes adaptive, guaranteeing small tracking errors and bounded parameter estimates. Formal nonlinear stability proofs are given to show that the tracking error is small. The NN saturation and FL deadzone compensator is simulated on a robot system to show its efficacy.