• Journal of the Korean Institute of Telematics and Electronics C
• /
• v.35C no.2
• /
• pp.56-64
• /
• 1998

Multi-layered perceptrons that are a nonlinear neural network model, have been widely used for various applications mainly thanks to good function approximation capability for nonlinear fuctions. However, for digital hardware implementation of the multi-layere perceptrons, the quantization scheme using "look-up tables (LUTs)" is commonly employed to handle nonlinear signmoid activation functions in the neworks, and thus requires large amount of storage to prevent unacceptable quantization errors. This paper is concerned with a new effective methodology for digital hardware implementation of multi-layered perceptrons, and proposes a "piecewise affine approximation" method in which input domain is divided into (small number of) sub-intervals and nonlinear sigmoid function is linearly approximated within each sub-interval. Using the proposed method, we develop an expression and an error backpropagation type learning algorithm for a multi-layered perceptron, and compare the performance with the quantization method through Monte Carlo simulations on XOR problems. Simulation results show that, in terms of learning convergece, the proposed method with a small number of sub-intervals significantly outperforms the quantization method with a very large storage requirement. We expect from these results that the proposed method can be utilized in digital system implementation to significantly reduce the storage requirement, quantization error, and learning time of the quantization method.quantization method.

• Journal of the Korean Data and Information Science Society
• /
• v.14 no.4
• /
• pp.825-836
• /
• 2003

Based on customer information and financing processes in capital market, we derived individual models by applying multi-layered perceptrons, MDA, and decision tree. Further, the results from the existing single models were compared with the results from the integrated model that was developed using genetic algorithm. This study contributes not only to verifying the existing individual models and but also to overcoming the limitations of the existing approaches. We have depended upon the approaches that compare individual models and search for the best-fit model. However, this study presents a methodology to build an integrated data mining model using genetic algorithm.

• The KIPS Transactions:PartD
• /
• v.12D no.2 s.98
• /
• pp.211-218
• /
• 2005

This study presents an integrated data mining model for the credit evaluation of the customers of a capital company. Based on customer information and financing processes in capital market, we derived individual models from multi-layered perceptrons(MLP), multivariate discrimination analysis(MDA), and decision tree. Further, the results from the existing models were compared with the results from the integrated model using genetic algorithm. The integrated model presented by this study turned out to be superior to the existing models. This study contributes not only to verifying the existing individual models but also to overcoming the limitations of the existing approaches.

• Journal of Korea Multimedia Society
• /
• v.13 no.3
• /
• pp.438-456
• /
• 2010

Iris recognition uses a unique iris pattern of user to identify person. In order to enhance the performance of iris recognition, it is reported that the diameter of iris region should be greater than 200 pixels in the captured iris image. So, the previous iris system used zoom lens camera, which can increase the size and cost of system. To overcome these problems, we propose a new method of enhancing the accuracy of iris recognition on low-resolution iris images which are captured without a zoom lens. This research is novel in the following two ways compared to previous works. First, this research is the first one to analyze the performance degradation of iris recognition according to the decrease of the image resolution by excluding other factors such as image blurring and the occlusion of eyelid and eyelash. Second, in order to restore a high-resolution iris image from single low-resolution one, we propose a new method based on multiple multi-layered perceptrons (MLPs) which are trained according to the edge direction of iris patterns. From that, the accuracy of iris recognition with the restored images was much enhanced. Experimental results showed that when the iris images down-sampled by 6% compared to the original image were restored into the high resolution ones by using the proposed method, the EER of iris recognition was reduced as much as 0.133% (1.485% - 1.352%) in comparison with that by using bi-linear interpolation

• Proceedings of the Korean Operations and Management Science Society Conference
• /
• 2001.10a
• /
• pp.161-164
• /
• 2001

본 연구에서는 할부금융시장에서의 고객신용예측을 위한 모형으로 여러 가지 인공신경망(Neural Network) 모형들을 유전자 알고리즘(Genetic Algorithm)을 이용하여 통합한 신용예측모형을 제안한다. 10개의 학습된 인공신경망 모형들을 유전자알고리즘을 이용하여 종류별로 통합하여 MLP(Multi-Layered Perceptrons), Linear, RBF(Radial Basis Function) 세 가지의 대표모델을 얻고 이를 다시 하나의 인공신경망 모델로 통합하였다. 이를 통합되기 이전의 각각의 인공신경망 모형들과 성능을 비교, 분석하여 본 연구에서 제안한 통합모형의 유효성과 통합방법의 타당성을 제시하였다.

• Journal of Information Technology Applications and Management
• /
• v.11 no.1
• /
• pp.161-174
• /
• 2004

This study is intended to suggest1 the optimized data mining model for the efficient customer credit evaluation in the capital finance industry. To accomplish the research objective, various data mining models for the customer credit evaluation are compared and analyzed. Furthermore, existing models such as Multi-Layered Perceptrons, Multivariate Discrimination Analysis, Radial Basis Function, Decision Tree, and Logistic Regression are employed for analyzing the customer information in the capital finance market and the detailed data of capital financing transactions. Finally, the data from the integrated model utilizing a genetic algorithm is compared with those of each individual model mentioned above. The results reveals that the integrated model is superior to other existing models.

• Nuclear Engineering and Technology
• /
• v.54 no.8
• /
• pp.3027-3033
• /
• 2022

Background: In this study, various types of deep-learning models for predicting in vitro radiosensitivity from gene-expression profiling were compared. Methods: The clonogenic surviving fractions at 2 Gy from previous publications and microarray gene-expression data from the National Cancer Institute-60 cell lines were used to measure the radiosensitivity. Seven different prediction models including three distinct multi-layered perceptrons (MLP), four different convolutional neural networks (CNN) were compared. Folded cross-validation was applied to train and evaluate model performance. The criteria for correct prediction were absolute error < 0.02 or relative error < 10%. The models were compared in terms of prediction accuracy, training time per epoch, training fluctuations, and required calculation resources. Results: The strength of MLP-based models was their fast initial convergence and short training time per epoch. They represented significantly different prediction accuracy depending on the model configuration. The CNN-based models showed relatively high prediction accuracy, low training fluctuations, and a relatively small increase in the memory requirement as the model deepens. Conclusion: Our findings suggest that a CNN-based model with moderate depth would be appropriate when the prediction accuracy is important, and a shallow MLP-based model can be recommended when either the training resources or time are limited.

• The Korean Journal of Nuclear Medicine
• /
• v.38 no.3
• /
• pp.233-240
• /
• 2004

Purpose: We developed a computer-aided classifier using artificial neural network (ANN) to discriminate the cerebral metabolic pattern of medial and lateral temporal lobe epilepsy (TLE). Materials and Methods: We studied brain F-18-FDG PET images of 113 epilepsy patients sugically and pathologically proven as medial TLE (left 41, right 42) or lateral TLE (left 14, right 16). PET images were spatially transformed onto a standard template and normalized to the mean counts of cortical regions. Asymmetry indices for predefined 17 mirrored regions to hemispheric midline and those for medial and lateral temporal lobes were used as input features for ANN. ANN classifier was composed of 3 independent multi-layered perceptrons (1 for left/right lateralization and 2 for medial/lateral discrimination) and trained to interpret metabolic patterns and produce one of 4 diagnoses (L/R medial TLE or L/R lateral TLE). Randomly selected 8 images from each group were used to train the ANN classifier and remaining 51 images were used as test sets. The accuracy of the diagnosis with ANN was estimated by averaging the agreement rates of independent 50 trials and compared to that of nuclear medicine experts. Results: The accuracy in lateralization was 89% by the human experts and 90% by the ANN classifier Overall accuracy in localization of epileptogenic zones by the ANN classifier was 69%, which was comparable to that by the human experts (72%). Conclusion: We conclude that ANN classifier performed as well as human experts and could be potentially useful supporting tool for the differential diagnosis of TLE.