• Journal of information and communication convergence engineering
• /
• v.9 no.6
• /
• pp.637-640
• /
• 2011

The hybrid neural networks have characteristics such as fast learning times, generality, and simplicity, and are mainly used to classify learning data and to model non-linear systems. The middle layer of a hybrid neural network clusters the learning vectors by grouping homogenous vectors in the same cluster. In the clustering procedure, the homogeneity between learning vectors is represented as the distance between the vectors. Therefore, if the distances between a learning vector and all vectors in a cluster are smaller than a given constant radius, the learning vector is added to the cluster. However, the usage of a constant radius in clustering is the primary source of errors and therefore decreases the recognition success rate. To improve the recognition success rate, we proposed the enhanced hybrid network that organizes the middle layer effectively by using the enhanced ART1 network adjusting the vigilance parameter dynamically according to the similarity between patterns. The results of experiments on a large number of calling card images showed that the proposed algorithm greatly improves the character extraction and recognition compared with conventional recognition algorithms.

• The Journal of the Acoustical Society of Korea
• /
• v.22 no.1E
• /
• pp.3-10
• /
• 2003

This paper describes two kinds of neural networks for stereoscopic vision, which have been applied to an identification of human speech. In speech recognition based on the stereoscopic vision neural networks (SVNN), the similarities are first obtained by comparing input vocal signals with standard models. They are then given to a dynamic process in which both competitive and cooperative processes are conducted among neighboring similarities. Through the dynamic processes, only one winner neuron is finally detected. In a comparative study, with, the average phoneme recognition accuracy on the two-layered SVNN was 7.7% higher than the Hidden Markov Model (HMM) recognizer with the structure of a single mixture and three states, and the three-layered was 6.6% higher. Therefore, it was noticed that SVNN outperformed the existing HMM recognizer in phoneme recognition.

• Smart Structures and Systems
• /
• v.25 no.4
• /
• pp.459-469
• /
• 2020

The presence of cracks in mechanical components is a very important problem that, if it is not detected on time, can lead to high economic costs and serious personal injuries. This work presents a methodology focused on identifying cracks in unbalanced rotors, which are some of the most frequent mechanical elements in industry. The proposed method is based on Artificial Neural Networks that give a solution to the presented inverse problem. They allow to estimate unknown crack parameters, specifically, the crack depth and the eccentricity angle, depending on the dynamic behavior of the rotor. The necessary data to train the developed Artificial Neural Network have been obtained from the frequency spectrum of the displacements of the well- known cracked Jeffcott rotor model, which takes into account the crack breathing mechanism during a shaft rotation. The proposed method is applicable to any rotating machine and it could contribute to establish adequate maintenance plans.

• Proceedings of the KIEE Conference
• /
• 2002.11d
• /
• pp.94-96
• /
• 2002

This paper is deal with the method of redesign for optimum thrust model using Neural-Networks in Permanent Magnet Linear Synchrous Motor(PMLSM). This method is saved time compared with design method using only Finite Element Method(FEM). In this paper data sets for training Neural-Networks obtained using 2D FEM. To confirm the validity of the data sets for training Neural-Networks optimum values of that Is compared with results of FEM. And then. this method is verified that it could be applied to the design for Slotless type PMLSM.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2000.05a
• /
• pp.114-117
• /
• 2000

In this paper, Fuzzy Self-organizing Neural Networks(FSONN) based on Fuzzy Neural Networks(FNN) is proposed to overcome some problems, such as the conflict between ovefitting and good generation, and low reliability. The proposed FSONN consists of FNN and SONN. Here, FNN is used as the premise part of FSONN and SONN is the consequnt part of FSONN. The FUN plays the preceding role of FSONN. For the fuzzy reasoning and learning method in FNN, Simplified fuzzy reasoning and backpropagation learning rule are utilized. The number of layers and the number of nodes in each layers of SONN that is based on the GMDH method are not predetermined, unlike in the case of the popular multi layer perceptron structure and can be generated. Also the partial descriptions of nodes can use various forms such as linear, modified quadratic, cubic, high-order polynomial and so on. In this paper, the optimal design procedure of the proposed FSONN is shown in each step and performance index related to approximation and generalization capabilities of model is evaluated and also discussed.

• Proceedings of the KIEE Conference
• /
• 1995.07b
• /
• pp.606-609
• /
• 1995

This paper presents discriminate between magnetizing inrush and internal faults of power transformer by artificial neural networks trained with preprocessing of fault discriminant. The proposed neural networks contain multi-layer perceptron using back-propagation learning algorithm with logistic sigmoid activation function. For this training and test, we used the relaying signals obtained from the EMTP simulation of model power system. It is shown that the proposed transformer protection system by neural networks never misoperated.

• The Journal of Korea Robotics Society
• /
• v.17 no.2
• /
• pp.133-141
• /
• 2022

This paper introduces model compression algorithms which make a deep neural network smaller and faster for embedded systems. The model compression algorithms can be largely categorized into pruning, quantization and knowledge distillation. In this study, gradual pruning, quantization aware training, and knowledge distillation which learns the activation boundary in the hidden layer of the teacher neural network are integrated. As a large deep neural network is compressed and accelerated by these algorithms, embedded computing boards can run the deep neural network much faster with less memory usage while preserving the reasonable accuracy. To evaluate the performance of the compressed neural networks, we evaluate the size, latency and accuracy of the deep neural network, DenseNet201, for image classification with CIFAR-10 dataset on the NVIDIA Jetson Xavier.

• The Transactions of the Korean Institute of Electrical Engineers
• /
• v.39 no.10
• /
• pp.1121-1232
• /
• 1990

A hierarchical neural network structure is described for robot PTP trajectory planning. In the first level, the multi-layered Perceptron neural network is used for the inverse kinematics with the back-propagation learning procedure. In the second level, a saccade generation model based joint trajectory planning model in proposed and analyzed with several features. Various simulations are performed to investigate the characteristics of the proposed neural networks.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.9
• /
• pp.3583-3597
• /
• 2020

In this paper, a parallel deep learning model using a convolutional neural network and a dilated convolutional neural network is proposed to classify Alzheimer's disease with high accuracy in PET/CT images. The developed model consists of two pipelines, a conventional CNN pipeline, and a dilated convolution pipeline. An input image is sent through both pipelines, and at the end of both pipelines, extracted features are concatenated and used for classifying Alzheimer's disease. Complimentary abilities of both networks provide better overall accuracy than single conventional CNNs in the dataset. Moreover, instead of performing binary classification, the proposed model performs three-class classification being Alzheimer's disease, mild cognitive impairment, and normal control. Using the data received from Dong-a University, the model performs classification detecting Alzheimer's disease with an accuracy of up to 95.51%.

• The Korean Journal of Applied Statistics
• /
• v.37 no.3
• /
• pp.297-309
• /
• 2024

This paper evaluates the performance of the neural network autoregressive model combined with an exponential smoothing model, called the NNARX+ETS model. The combined model utilizes the components of ETS as exogenous variables for NNARX, to forecast time series data using artificial neural networks. The main idea is to enhance the performance of NNAR using only lags of the original time series data, by combining traditional time series analysis methods with the neural networks through NNARX. We employ two real data for performance evaluation and compare the NNARX+ETS with NNAR and traditional time series analysis methods such as ETS and ARIMA (autoregressive integrated moving average) models.