• Communications for Statistical Applications and Methods
• /
• v.4 no.1
• /
• pp.91-99
• /
• 1997

본 연구에서는 신경망이론을 이용하여 시계열자료를 분석할 때 문제가 되고 있는 초기 가중값을 선정하는 방법을 제시하고자 한다. 기존의 연구에서 학습을 위한 초기 가중값의 결정은 난수에 의존하고 있다. 본 연구에서는 신경망학습의 효율적인 초기값을 선택하기 위하여 제어상자를 이용한다. 그리고 학습과정에서 가중값의 변화를 추적하고 적절한 가중값의 범위를 탐색하면서 새로운 초기값을 제어상자를 통하여 실시간으로 재설정하는 방법을 제시한다.

• The Korean Journal of Applied Statistics
• /
• v.32 no.3
• /
• pp.375-389
• /
• 2019

Sample attrition according to a long-term tracking reduces the representativeness of the sample data in a panel study. Most panel surveys in South Korea and other countries have prepared response adjustment weights in order to solve problems regarding representativeness due to sample attrition. In this paper, we divided the panel data into continuous response group and non-continuous response group according to response patterns and considered a weighting adjustment method to reduce the bias of the non-continuous response group. A simulation indicated that the proposed composite estimation type weighting method, which reflected the characteristics of non-continuous response groups, could be more efficient than other weighting methods in terms of reducing non-response bias. As a case study, the proposed methods are applied to the Korean Longitudinal Study of Ageing (KLoSA) data of the Korea Employment Information Service.

• The KIPS Transactions:PartB
• /
• v.10B no.7
• /
• pp.757-768
• /
• 2003

This paper proposes a weighted fuzzy reasoning algorithm for rule-based systems based on weighted fuzzy Pr/T nets, where the certainty factors of the fuzzy production rules, the truth values of the predicates appearing in the rules and the weights representing the importance of the predicates are represented by the fuzzy numbers. The proposed algorithm is more flexible and much closer to human intuition and reasoning than other methods : $\circled1$ calculate the certainty factors using by the simple min and max operations based on the only certainty factors of the fuzzy production rules without the weights of the predicates[10] : $\circled2$ evaluate the belief of the fuzzy production rules using by the belief evaluation functions according to fuzzy concepts in the fuzzy rules without the weights of the predicates[12], because this algorithm uses the weights representing the importance of the predicates in the fuzzy production rules.

• Proceedings of the IEEK Conference
• /
• 2003.07d
• /
• pp.1275-1278
• /
• 2003

본 논문에서는 오차 역전파 알고리즘의 전역 최소값을 찾지 못하는 문제점에 대해서 설명하였고, 이 문제를 해결하기 위한 방법으로 유전자 알고리즘에 대해서 설명하였다. 오차 역전파 알고리즘은 기본적으로 경도 하강법을 따른다. 따라서 신경망의 각 가중값 행렬이 만드는 고차의 오차 평면이 대부분의 문제에서 다수의 국부 최소값들을 가지는게 일반적인데, 가중값의 변화가 한방으로 진행하기 시작하여, 오차가 증가되어지는 언덕이 학습 계수보다 크다면 더 이상 학습은 진행되지 않고 거기에서 빠져나가지 못한다. 따라서 초기의 위치가 중요한 역할을 하는데, 이 문제를 해결하기 위해서 유전자 알고리즘을 이용한 신경망 초기화 방법을 제안하였다. 끝으로, 간단한 실험으로 제안된 방법을 구현하고 결과에 대해서 논하였다

• The Korean Journal of Applied Statistics
• /
• v.26 no.5
• /
• pp.847-856
• /
• 2013

We propose a modified learning process for generalized neural network using a learning algorithm by Liu et al. (2001). We consider the effect of initial weights, training results and learning errors using a modified learning process. We employ an incremental training procedure where training patterns are learned systematically. Our algorithm starts with a single training pattern and a single hidden layer neuron. During the course of neural network training, we try to escape from the local minimum by using a weight scaling technique. We allow the network to grow by adding a hidden layer neuron only after several consecutive failed attempts to escape from a local minimum. Our optimization procedure tends to make the network reach the error tolerance with no or little training after the addition of a hidden layer neuron. Simulation results with suitable initial weights indicate that the present constructive algorithm can obtain neural networks very close to minimal structures and that convergence to a solution in neural network training can be guaranteed. We tested these algorithms extensively with small training sets.

• Journal of Internet Computing and Services
• /
• v.8 no.3
• /
• pp.99-107
• /
• 2007

In general, the reliabilities of the fuzzy system are represented and analyzed by real numbers between zero and one, fuzzy numbers, intervals of confidence, interval-valued fuzzy sets, vague sets, etc. This paper addresses the method to analyze the reliability of the fuzzy system for the weighted components with the weights reflected on the importance of weighted components in an system. The reliabilities and the weights of the weighted components in a fuzzy numbers and considers the weights of the weighted components in a fuzzy system, therefore, its execution is faster and more flexible than the conventional methods.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2004.11a
• /
• pp.493-496
• /
• 2004

본 연구에서는 신경망 학습의 중요한 평가 척도로써 고려될 수 있는 일반화 성능과 학습속도를 개선시키기 위한 방안으로써 초기 가중값과 학습률과 같은 주요 인자들을 이용한 신경망 학습 영향을 살펴본다. 특히 초기 가중값과 학습률을 고정시킨 후 새롭게 조정된 계수들을 점차적으로 변화시키는 새로운 인자 결합방법을 이용하여 신경망 학습량과 학습속도를 비교해 보고 계수조정을 통한 개선된 학습 영향을 살펴본다. 그리고 단순한 예제를 이용한 실증분석을 통하여 신경망 모형의 일반화 성능과 학습 속도 개선을 위한 각 인자들의 개별 효과와 결합 효과를 살펴보고 그 개선 방안을 제시한다.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.34 no.3
• /
• pp.795-804
• /
• 2014

In this study, Andong Reservoir monthly and ten days inflows from July 2011 to September 2011 are predicted using SWAT model and ensemble technique. The weight method using monthly and ten days rainfall forecasts from Korea Meteorological Administration is applied for accurate analysis. If the rainfall prediction announced by Korea Meteorological Administration is close to the actual rainfall, the PDF-Ratio Method shows the best result. If the past high rainfall occurrence is close to the actual rainfall, the modified PDF-Ratio method shows the best result. This method can improve the prediction accuracy even though the Korea Meteorological Administration forecast is not accurate. On the contrary, if Korea Meteorological Administration forecast is different from the actual rainfall and the past rainfall occurrence statistics of lower section, the uniform method shows the best result.

• Journal of Internet Computing and Services
• /
• v.5 no.4
• /
• pp.115-124
• /
• 2004

This paper presents a weighted fuzzy backward reasoning algorithm for rule-based systems based on weighted fuzzy Petri nets. The fuzzy production rules in the knowledge base of a rule-based system are modeled by weighted fuzzy Petri nets, where the truth values of the propositions appearing in the fuzzy production rules and the certainty factors of the rules are represented by fuzzy numbers. Furthermore, the weights of the propositions appearing in the rules are also represented by fuzzy numbers. The proposed weighted fuzzy backward reasoning generates the backward reasoning path from the goal node to the initial nodes and then evaluates the certainty factor of the goal node. The algorithm we proposed can allow the rule-based systems to perform weighted fuzzy backward reasoning in more flexible and human-like manner.

• Journal of KIISE:Software and Applications
• /
• v.33 no.11
• /
• pp.979-985
• /
• 2006

In the conventional researches, the reliabilities of the fuzzy system are represented and analyzed by real values between zero and one, fuzzy numbers, intervals of confidence, etc. In this paper, we present a method to represent and analyze the reliabilities of the weighted components of the fuzzy system and the weights reflected on their importance based on vague sets defined in the universe of discourse [0, 1]. The vague set is represented as the interval consisted of the truth-membership functions and the false-membership functions, therefore it can allow the reliabilities and the weights of a fuzzy system to represent in a more flexible manner. The proposed method considers the weights of the weighted components in the fuzzy systems, its reliability analysis is more flexible and effective than the conventional methods.