• 한국컴퓨터정보학회논문지
• /
• 제23권11호
• /
• pp.43-49
• /
• 2018

In this paper, we propose a displacement measurement method based on deep learning using image data obtained from tensile tests of a material specimen. We focus on the fact that the sequential images during the tension are generated and the displacement of the specimen is represented in the image data. So, we designed sample generation model which makes sequential images of specimen. The behavior of generated images are similar to the real specimen images under tensile force. Using generated images, we trained and validated our model. In the deep neural network, sequential images are assigned to a multi-channel input to train the network. The multi-channel images are composed of sequential images obtained along the time domain. As a result, the neural network learns the temporal information as the images express the correlation with each other along the time domain. In order to verify the proposed method, we conducted experiments by comparing the deformation measuring performance of the neural network changing the displacement range of images.

• Journal of Platform Technology
• /
• 제6권4호
• /
• pp.34-40
• /
• 2018

현재 인공신경망은 단일 작업에 대해선 뛰어난 성능을 보이나, 다른 종류의 작업을 학습하면 이전 학습 내용을 잊어버리는 단점이 있다. 이를 catastrophic forgetting이라고 한다. 인공신경망의 활용도를 높이긴 위해선 이 현상을 극복해야 한다. catastrophic forgetting을 극복하기 위한 여러 노력이 있다. 하지만 많은 노력이 있었음에도 완벽하게 catastrophic forgetting을 극복하지는 못하였다. 본 논문에서는 여러 노력 중 elastic weight consolidation(EWC)에 사용되는 핵심 개념을 이용하여, 순차적 반복학습을 제시한다. 인공신경망 학습에 많이 쓰이는 MNIST를 확장한 EMNIST 데이터 셋을 이용하여 catastrophic forgetting 현상을 재현하고 이를 순차적 반복학습을 통해 극복하는 실험을 진행하였으며, 그 결과 모든 작업에 대해서 학습이 가능하였다.

• 대한기계학회논문집A
• /
• 제32권3호
• /
• pp.258-266
• /
• 2008

The conventional approximate optimization method, which uses the statistical design of experiments(DOE) and response surface method(RSM), can derive an approximated optimum results through the iterative process by a trial and error. The quality of results depends seriously on the factors and levels assigned by a designer. The purpose of this study is to propose a new technique, which is called a sequential design of experiments(SDOE), to reduce a trial and error procedure and to find an appropriate condition for using artificial neural network(ANN) systematically. An appropriate condition is determined from the iterative process based on the analysis of means. With this new technique and ANN, it is possible to find an optimum design accurately and efficiently. The suggested algorithm has been applied to various mathematical examples and a structural problem.

• 대한인간공학회지
• /
• 제21권1호
• /
• pp.1-14
• /
• 2002

The main purpose of this paper is to present selection criteria for ROK Airforce pilot training candidates in order to save costs involved in sequential pilot training. We use classification models such Decision Tree, Logistic Regression and Neural Network based on aptitude test results of 288 ROK Air Force applicants in 1994-1996. Different models are compared in terms of classification accuracy, ROC and Lift-value. Neural network is evaluated as the best model for each sequential flight test result while Logistic regression model outperforms the rest of them for discriminating the last flight test result. Therefore we suggest a pilot selection criterion based on this logistic regression. Overall. we find that the factors such as Attention Sharing, Speed Tracking, Machine Comprehension and Instrument Reading Ability having significant effects on the flight results. We expect that the use of our criteria can increase the effectiveness of flight resources.

• Communications for Statistical Applications and Methods
• /
• 제9권1호
• /
• pp.155-166
• /
• 2002

Neural networks have been studied as a popular tool for classification and they are very flexible. Also, they are used for many applications of pattern classification and pattern recognition. This paper focuses on Bayesian approach to feed-forward neural networks with single hidden layer of units with logistic activation. In this model, we are interested in deciding the number of nodes of neural network model with p input units, one hidden layer with m hidden nodes and one output unit in Bayesian setup for fixed m. Here, we use the latent variable into the prior of the coefficient regression, and we introduce the 'sequential step' which is based on the idea of the data augmentation by Tanner and Wong(1787). The MCMC method(Gibbs sampler and Metropolish algorithm) can be used to overcome the complicated Bayesian computation. Finally, a proposed method is applied to a simulated data.

• 한국멀티미디어학회논문지
• /
• 제23권2호
• /
• pp.174-185
• /
• 2020

Leukemia induced death has been listed in the top ten most dangerous mortality basis for human being. Some of the reason is due to slow decision-making process which caused suitable medical treatment cannot be applied on time. Therefore, good clinical decision support for acute leukemia type classification has become a necessity. In this paper, the author proposed a novel approach to perform acute leukemia type classification using sequential neural network classifier. Our experimental result only cover the first classification process which shows an excellent performance in differentiating normal and abnormal cells. Further development is needed to prove the effectiveness of second neural network classifier.

• 응용통계연구
• /
• 제18권1호
• /
• pp.137-146
• /
• 2005

신경망은 전통적인 시계열 기법들에 비해 대체적으로 예측성능의 우수함이 입증되었으나 계절성과 추세성을 갖는 시계열자료에 대해 예측력이 떨어지는 단점을 가지고 있다. 최근에는 Ensemble 기법인 Bagging Algorithm과 신경망의 혼합모형인 Bagging Neural Network이 개밭되었다. 이 기법은 분산과 편향을 많이 줄여줌으로써 더 좋은 예측을 할 수 있는 것으로 나타났다. 그러나 Ensemble 기법을 이용한 예측모형은 시계열자료를 적합 시키는데 있어 초기부여확률 및 예측자 선정시의 문제점을 가지고 있다. 이에 본 연구에서는 이러한 문제점을 해결하고 더불어 예측력을 향상시키기 위한 방법으로 초기부여확률이 균일분포가 아닌 순차적인 형태의 분포를 제시하고 신경망을 예측자로 활용한 변형된 Ensemble Algorithm을 제안한다. 또한 예측모형의 평가를 위해 실제자료를 가지고 기존 예측모형들과 제안한 방법을 이용하여 예측하고 각 MSE의 비교를 통하여 예측정확도를 알아보고자 한다.

• Structural Engineering and Mechanics
• /
• 제51권6호
• /
• pp.989-1003
• /
• 2014

In the preliminary design stage of an RC 3D-frame, repeated sequential analyses to determine optimal members' sizes and the investigation of the parameters required to minimize the differential column shortening are computational effort consuming, especially when considering various types of loads such as dead load, temperature action, time dependent effects, construction and live loads. Because the desired accuracy at this stage does not justify such luxury, two backpropagation feedforward artificial neural networks have been proposed in order to approximate this information. Instead of using a commercial software package, many references providing advanced principles have been considered to code a program and generate these neural networks. The first one predicts the typical amount of time between two phases, needed to achieve the minimum maximorum differential column shortening. The other network aims to prognosticate sequential analysis results from those of the simultaneous analysis. After the training stages, testing procedures have been carried out in order to ensure the generalization ability of these respective systems. Numerical cases are studied in order to find out how good these ANN match with the sequential finite element analysis. Comparison reveals an acceptable fit, enabling these systems to be safely used in the preliminary design stage.

• International Journal of Advanced Culture Technology
• /
• 제8권2호
• /
• pp.231-238
• /
• 2020

Generally, sequential data refers to data having continuity. Text data, which is a representative type of unstructured data, is also sequential data in that it is necessary to know the meaning of the preceding word in order to know the meaning of the following word or context. So far, many techniques for analyzing sequential data such as text data have been proposed. In this paper, four methods of 1d-CNN, LSTM, BiLSTM, and C-LSTM are introduced, focusing on neural network techniques. In addition, by using this, IMDb movie review data was classified into two classes to compare the performance of the techniques in terms of accuracy and analysis time.

• 인지과학
• /
• 제13권1호
• /
• pp.13-21
• /
• 2002

퍼지 논리를 이용한 추론은 일부의 정보가 무시되어 적절하지 못한 추론 결과를 초래할 수 있다. 또한 신경망은 패턴 처리에는 적합하지만 인간의 지식을 모델링하기 위해서 필요한 논리적인 추론에는 부적합하다. 하지만 신경 망의 변형인 신경 논리 망을 이용하면 논리적인 추론이 가능하다. 따라서 본 논문에서는 기존의 신경 논리 망을 기반으로 하는 추론 망을 확장하여 퍼지 추론 망을 구성하고 기존의 추론 망에서 사용되는 전파규칙을 보완하여 적용하고자 한다. 퍼지 추론 망에서 퍼지 규칙의 결론부에 해당하는 명제의 믿음 값을 결정하기 위해서는 추론하고자 하는 명제에 연결된 노드들을 탐색해야 한다. 이를 위해, 연결된 모든 노드들의 링크를 따라 순차적인 탐색을 하는 경우와 링크에 부여된 우선순위에 의해 탐색을 하는 경우의 탐색비용에 대하여 실험을 통해 비교 평가하였다. 실험결과 퍼지 추론 망의 크기가 확장될수록, 그리고 탐색 경험의 횟수가 증가할수록 순차적인 탐색전략보다 우선순위에 의한 탐색전략이 탐색 비용면에서 효율성이 더욱 증가함을 알 수 있었다.