• 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.

• Food Science of Animal Resources
• /
• v.38 no.2
• /
• pp.362-375
• /
• 2018

This study proposed a rapid microscopic examination method for pork freshness evaluation by using the self-assembled hyperspectral microscopic imaging (HMI) system with the help of feature extraction algorithm and pattern recognition methods. Pork samples were stored for different days ranging from 0 to 5 days and the freshness of samples was divided into three levels which were determined by total volatile basic nitrogen (TVB-N) content. Meanwhile, hyperspectral microscopic images of samples were acquired by HMI system and processed by the following steps for the further analysis. Firstly, characteristic hyperspectral microscopic images were extracted by using principal component analysis (PCA) and then texture features were selected based on the gray level co-occurrence matrix (GLCM). Next, features data were reduced dimensionality by fisher discriminant analysis (FDA) for further building classification model. Finally, compared with linear discriminant analysis (LDA) model and support vector machine (SVM) model, good back propagation artificial neural network (BP-ANN) model obtained the best freshness classification with a 100 % accuracy rating based on the extracted data. The results confirm that the fabricated HMI system combined with multivariate algorithms has ability to evaluate the fresh degree of pork accurately in the microscopic level, which plays an important role in animal food quality control.

• Journal of Korean Society of Steel Construction
• /
• v.17 no.6 s.79
• /
• pp.661-672
• /
• 2005

For the efficient stochastic optimization of steel structures for which a large number of analyses is required, artificial neural networks,which have emerged as a powerful tool that could have been used to replace time-consuming procedures in many scientific or engineering applications, are applied. They are utilized for the solution of the equilibrium equations resulting from the application of the finite element method in connection with the reanalysis type of problem, for which a large number of finite element analyses are required in this study. As such, the use of artificial neural networks to predict finite element analysis outputs simplifies and facilitates the performance of the stochastic optimal design of structural systems where a trained neural network is used to replace the structural reanalysis phase. Moreover, to improve efficiency of used artificial neural networks, genetic algorithm is utilized. The stochastic optimizer used in this study is an algorithm based on the evolution theory. The efficiency of the proposed procedure is examined in problems with both volume (weight) functions and real-world cost functions

• The Journal of the Acoustical Society of Korea
• /
• v.13 no.4
• /
• pp.60-68
• /
• 1994

In this paper we propose a new Hybrid learning rule applied to multilayer feedforward neural networks, which is constructed by combining Hebbian learning rule that is a good feature extractor and Back-Propagation(BP) learning rule that is an excellent classifier. Unlike the BP rule used in multi-layer perceptron(MLP), the proposed Hybrid learning rule is used for uptate of all connection weights except for output connection weigths becase the Hebbian learning in output layer does not guarantee learning convergence. To evaluate the performance, the proposed hybrid rule is applied to classifier problems in two dimensional space and shows better performance than the one applied only by the BP rule. In terms of learning speed the proposed rule converges faster than the conventional BP. For example, the learning of the proposed Hybrid can be done in 2/10 of the iterations that are required for BP, while the recognition rate of the proposed Hybrid is improved by about $0.778\%$ at the peak.

• Journal of Sensor Science and Technology
• /
• v.7 no.6
• /
• pp.418-425
• /
• 1998

Stereo cameras are the most widely used sensing systems for automated machines including robots to interact with their three-dimensional(3D) working environments. The position of a target point in the 3D world coordinates can be measured by the use of stereo cameras and the camera calibration is an important preliminary step for the task. Existing camera calibration techniques can be classified into two large categories - linear and nonlinear techniques. While linear techniques are simple but somewhat inaccurate, the nonlinear ones require a modeling process to compensate for the lens distortion and a rather complicated procedure to solve the nonlinear equations. In this paper, a method employing a neural network for the calibration problem is described for tackling the problems arisen when existing techniques are applied and the results are reported. Particularly, it is shown experimentally that by utilizing the function approximation capability of multi-layer neural networks trained by the back-propagation(BP) algorithm to learn the error pattern of a linear technique, the measurement accuracy can be simply and efficiently increased.

• The Transactions of the Korean Institute of Electrical Engineers D
• /
• v.49 no.7
• /
• pp.378-388
• /
• 2000

In this paper, we propose Fuzzy Polynomial Neural Networks(FPNN) based on Polynomial Neural Networks(PNN) and Fuzzy Neural Networks(FNN) for model identification of complex and nonlinear systems. The proposed FPNN is generated from the mutually combined structure of both FNN and PNN. The one and the other are considered as the premise part and consequence part of FPNN structure respectively. As the consequence part of FPNN, PNN is based on Group Method of Data Handling(GMDH) method and its structure is similar to Neural Networks. But the structure of PNN is not fixed like in conventional Neural Networks and self-organizing networks that can be generated. FPNN is available effectively for multi-input variables and high-order polynomial according to the combination of FNN with PNN. Accordingly it is possible to consider the nonlinearity characteristics of process and to get better output performance with superb predictive ability. As the premise part of FPNN, FNN uses both the simplified fuzzy inference as fuzzy inference method and error back-propagation algorithm as learning rule. The parameters such as parameters of membership functions, learning rates and momentum coefficients are adjusted using genetic algorithms. And we use two kinds of FNN structure according to the division method of fuzzy space of input variables. One is basic FNN structure and uses fuzzy input space divided by each separated input variable, the other is modified FNN structure and uses fuzzy input space divided by mutually combined input variables. In order to evaluate the performance of proposed models, we use the nonlinear function and traffic route choice process. The results show that the proposed FPNN can produce the model with higher accuracy and more robustness than any other method presented previously. And also performance index related to the approximation and prediction capabilities of model is evaluated and discussed.

• Proceedings of the KIEE Conference
• /
• 2005.11b
• /
• pp.145-147
• /
• 2005

This paper presents a methodology for satisfying the thermal comfort of Indoor environment and reducing the summer peak demand power by minimizing the power consumption for an Air-conditioner within a space. KEPCO(Korea Electric Power Corporation) use the fixed duty cycle control method regardless of the indoor thermal environment. This method has disadvantages that energy saying depends on the set-point value of the Air-Conditioner and DLC has no net effects on Air-conditioners if the appliance has a lower operating cycle than the fixed duty cycle. A variable duty cycle estimates the PMV(Predict Mean Vote) at the next step with a predicted temperature and humidity coming from the back propagation neural network model. It is possible to reduce the energy consumption by maintaining the Air-conditioner's OFF state when the PMV lies in the thermal comfort range. The proposed methodology uses the historical real data of Sep. 7th, 2001 from a classroom in seoul to verify the effectiveness of the variable duty cycle method comparing with fixed duty cycle. The result shows that the variable duty cycle reduces the peak demand to 2.6times more than fixed duty cycle and increases the load control ratio by 8% more. Based on the variable duty cycle control algorithm, the effectiveness of DLC is much more improved as compared with the fixed duty cycle.

• Korean Journal of Cognitive Science
• /
• v.9 no.1
• /
• pp.1-11
• /
• 1998

In this paper, the features which are thought to be used by humans based on the psychological experiment of human information processing are applied to character recognition problem. Man will deal with a little large area information as well as pixel by pixel information. Therefore we define the feature that represents a little wide region I information called region feature, and combine the features derived from region feature and pixel by pixel features that have been used by now. The features we used are the result of region feature based preanalysis, mesh with region attributes, cross distance difference and gradient. The training and test data in the experiment are handwritten Korean alphabets, digits and English alphabets, which are trained on neural network using back propagation algorithm and recognition results are 90.27-93.25%, 98.00% and 79.73-85.75%, respectively Experimental results show that the feature we are suggesting in this paper is 1-2% better than UDLRH feature similar in attribute to region feature, and the tendency of misrecognition is more easily acceptable by humans.

• The Journal of Korean Association of Computer Education
• /
• v.6 no.2
• /
• pp.33-40
• /
• 2003

Criticality prediction models that determine whether a design entity is fault-prone or non fault-prone play an important role in reducing system development costs because the problems in early phases largely affect the quality of the late products. Real-time systems such as telecommunication systems are so large that criticality prediction is mere important in real-time system design. The current models are based on the technique such as discriminant analysis, neural net and classification trees. These models have some problems with analyzing causes of the prediction results and low extendability. This paper builds a new prediction model, GAM, based on Genetic Algorithm. GAM is different from other models because it produces a criticality function. So GAM can be used for comparison between entities by criticality. GAM is implemented and compared with a well-known prediction model, BackPropagation neural network Model(BPM), considering Internal characteristics and accuracy of prediction.

• The Journal of Korean Association of Computer Education
• /
• v.8 no.5
• /
• pp.129-137
• /
• 2005

Criticality prediction models that identify fault-prone spots using system design specifications play an important role in reducing development costs of large systems such as telecommunication systems. Many criticality prediction models using complexity metrics have been suggested. But most of them need training data set for model training. And they are classification models that can only classify design entities into fault-prone group and non fault-prone group. To solve this problem, this paper builds a new prediction model, HMM, using two styled hybrid metrics. HMM has strong point that it does not need training data and it enables comparison between design entities by criticality. HMM is implemented and compared with a well-known prediction model, BackPropagation neural network Model(BPM), considering internal characteristics and accuracy of prediction.