• Journal of Korea Multimedia Society
• /
• v.24 no.8
• /
• pp.979-987
• /
• 2021

Multiple Sclerosis (MS) can be early diagnosed by detecting lesions in brain magnetic resonance images (MRI). Unsupervised anomaly detection methods based on autoencoder have been recently proposed for automated detection of MS lesions. However, these autoencoder-based methods were developed only for 2D images (e.g. 2D cross-sectional slices) of MRI, so do not utilize the full 3D information of MRI. In this paper, therefore, we propose a novel 3D autoencoder-based framework for detection of the lesion volume of MS in MRI. We first define a 3D convolutional neural network (CNN) for full MRI volumes, and build each encoder and decoder layer of the 3D autoencoder based on 3D CNN. We also add a skip connection between the encoder and decoder layer for effective data reconstruction. In the experimental results, we compare the 3D autoencoder-based method with the 2D autoencoder models using the training datasets of 80 healthy subjects from the Human Connectome Project (HCP) and the testing datasets of 25 MS patients from the Longitudinal multiple sclerosis lesion segmentation challenge, and show that the proposed method achieves superior performance in prediction of MS lesion by up to 15%.

• The Korean Journal of Applied Statistics
• /
• v.32 no.5
• /
• pp.693-702
• /
• 2019

Deep learning has gained popularity for the classification and prediction task. Neural network layers become deeper as more data becomes available. Saturation is the phenomenon that the gradient of an activation function gets closer to 0 and can happen when the value of weight is too big. Increased importance has been placed on the issue of saturation which limits the ability of weight to learn. To resolve this problem, Glorot and Bengio (Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, 249-256, 2010) claimed that efficient neural network training is possible when data flows variously between layers. They argued that variance over the output of each layer and variance over input of each layer are equal. They proposed a method of initialization that the variance of the output of each layer and the variance of the input should be the same. In this paper, we propose a new method of establishing initialization by adopting truncated normal distribution and truncated cauchy distribution. We decide where to truncate the distribution while adapting the initialization method by Glorot and Bengio (2010). Variances are made over output and input equal that are then accomplished by setting variances equal to the variance of truncated distribution. It manipulates the distribution so that the initial values of weights would not grow so large and with values that simultaneously get close to zero. To compare the performance of our proposed method with existing methods, we conducted experiments on MNIST and CIFAR-10 data using DNN and CNN. Our proposed method outperformed existing methods in terms of accuracy.

• The Transactions of the Korea Information Processing Society
• /
• v.4 no.7
• /
• pp.1749-1758
• /
• 1997

This paper suggests the development of dynamic forecasting model for short-term power demand based on Radial Basis Function Network and Pal's GLVQ algorithm. Radial Basis Function methods are often compared with the backpropagation training, feed-forward network, which is the most widely used neural network paradigm. The Radial Basis Function Network is a single hidden layer feed-forward neural network. Each node of the hidden layer has a parameter vector called center. This center is determined by clustering algorithm. Theatments of classical approached to clustering methods include theories by Hartigan(K-means algorithm), Kohonen(Self Organized Feature Maps %3A SOFM and Learning Vector Quantization %3A LVQ model), Carpenter and Grossberg(ART-2 model). In this model, the first approach organizes the load pattern into two clusters by Pal's GLVQ clustering algorithm. The reason of using GLVQ algorithm in this model is that GLVQ algorithm can classify the patterns better than other algorithms. And the second approach forecasts hourly load patterns by radial basis function network which has been constructed two hidden nodes. These nodes are determined from the cluster centers of the GLVQ in first step. This model was applied to forecast the hourly loads on Mar. $4^{th},\;Jun.\;4^{th},\;Jul.\;4^{th},\;Sep.\;4^{th},\;Nov.\;4^{th},$ 1995, after having trained the data for the days from Mar. $1^{th}\;to\;3^{th},\;from\;Jun.\;1^{th}\;to\;3^{th},\;from\;Jul.\;1^{th}\;to\;3^{th},\;from\;Sep.\;1^{th}\;to\;3^{th},\;and\;from\;Nov.\;1^{th}\;to\;3^{th},$ 1995, respectively. In the experiments, the average absolute errors of one-hour ahead forecasts on utility actual data are shown to be 1.3795%.

• Proceedings of the KIEE Conference
• /
• 2000.07a
• /
• pp.212-214
• /
• 2000

The distance relay technique used for transmission line protection operates overreach and underreach to the self protection region because the power system becomes complex and fault conditions are different. To solve these problems, this paper describes new technique to set the reliable self protection lesion. The trip region of the quadrilateral distance relay is set by training of multi layer recurrent elman network. The proposed network is able to reach the trip zone for the fault impedance, fault initial angle and source impedance variance correctly.

• 제어로봇시스템학회:학술대회논문집
• /
• 2005.06a
• /
• pp.275-280
• /
• 2005

The principal context of this research is the approach to an artificial neural network algorithm which solves multivariable nonlinear equation systems by estimating the state of line power flow. First a dynamical neural network with feedback is used to find the minimum value of the objective function at each iteration of the state estimator algorithm. In second step a two-layer neural network structures is derived to implement all of the different matrix-vector products that arise in neural network state estimator analysis. For hardware requirements, as they relate to the total number of internal connections, the architecture developed here preserves in its structure the pronounced sparsity of power networks for which state the estimator analysis is to be carried out. A principal feature of the architecture is that the computing time overheads in solution are independent of the dimensions or structure of the equation system. It is here where the ultrahigh-speed of massively parallel computing in neural networks can offer major practical benefit.

• Journal of information and communication convergence engineering
• /
• v.8 no.3
• /
• pp.281-288
• /
• 2010

This paper investigates the performance of IEEE 802.11p wireless access in vehicular environments (WAVE) system in small-scale fading models reported by Georgia Institute of Technology (Georgia Tech). We redesign the small-scale fading models to be applied to the computer simulation and develop the IEEE 802.11p WAVE physical layer simulator to provide the bit error rate and packet error rate performances. Moreover, a new channel estimator using decision directed channel estimation and linear minimum mean square error smoothing is proposed in order to improve the performance of the conventional least square channel estimator using two identical long training symbols. The simulation results are satisfactorily coincident with the scenarios of Georgia Tech report, and the proposed channel estimator significantly outperforms the conventional channel estimator.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.32B no.5
• /
• pp.756-764
• /
• 1995

In this paper, we proposed MLP(MultiLayer Perceptron) based HMM's(Hidden Markov Models) with bounded state duration for isolated word recognition. The minimum and maximum state duration for each state of a HMM are estimated during the training phase and used as parameters of constraining state transition in a recognition phase. The procedure for estimating these parameters and the recognition algorithm using the proposed HMM's are also described. Speaker independent isolated word recognition experiments using a vocabulary of 10 city names and 11 digits indicate that recognition rate can be improved by adjusting the minimum state durations.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.58 no.12
• /
• pp.2532-2537
• /
• 2009

For defect characterization in steam generator tubes in nuclear power plant, artificial neural network has been extensively used to classify defect types. In this paper, we study the effectiveness of Bagging for improving the performance of neural network for the classification of tube defects. Bagging is a method that combines outputs of many neural networks that were trained separately with different training data set. By varying the number of neurons in the hidden layer, we carry out computer simulations in order to compare the classification performance of bagging neural network and single neural network. From the experiments, we found that the performance of bagging neural network is superior to the average performance of single neural network in most cases.

• Journal of the Korean Society for Precision Engineering
• /
• v.16 no.10
• /
• pp.25-34
• /
• 1999

In this paper, an algorithm is presented to recognize lane and obstacles based on highway road image. The road images obtained by a video camera undergoes a pre-processing that includes filtering, edge detection, and identification of lanes. After this pre-processing, a part of image is grouped into 27 sub-windows and fed into a three-layer feed-forward neural network. The neural network is trained to indicate the road direction and the presence of absence of an obstacle. The proposed algorithm has been tested with the images different from the training images, and demonstrated its efficacy for recognizing lane and obstacles. Based on the test results, it can be said that the algorithm successfully combines the traditional image processing and the neural network principles towards a simpler and more efficient driver warning of assistance system

• Proceedings of the Korea Association of Information Systems Conference
• /
• 2000.11a
• /
• pp.36-44
• /
• 2000

The purpose of this study is to introduce a more efficient forecasting technique, which could help result the reduction of cost in removing the waste of airline in-flight meals. We will use a neural network approach known to many researchers as the “Outstanding Forecasting Technique”. We employed a multi-layer perceptron neural network using a backpropagation algorithm. We also suggested using other related information to improve the forecasting performances of neural networks. We divided the data into three sets, which are training data set, cross validation data set, and test data set. Time lag variables are still employed in our model according to the general view of time series forecasting. We measured the accuracy of our model by “Mean Square Error”(MSE). The suggested model proved most excellent in serving economy class in-flight meals. Forecasting the exact amount of meals needed for each airline could reduce the waste of meals and therefore, lead to the reduction of cost. Better yet, it could enhance the cost competition of each airline, keep the schedules on time, and lead to better service.