• Journal of KIISE
• /
• v.44 no.7
• /
• pp.674-679
• /
• 2017

In big-data environments wherein large amounts of text documents are produced daily, titles are very important clues that enable a prompt catching of the key ideas in documents; however, titles are absent for numerous document types such as blog articles and social-media messages. In this paper, a title-generation model for which sequence-to-sequence RNNs with attention and copying mechanisms are employed is proposed. For the proposed model, input sentences are encoded based on bi-directional GRU (gated recurrent unit) networks, and the title words are generated through a decoding of the encoded sentences with keywords that are automatically selected from the input sentences. Regarding the experiments with 93631 training-data documents and 500 test-data documents, the attention-mechanism performances are more effective (ROUGE-1: 0.1935, ROUGE-2: 0.0364, ROUGE-L: 0.1555) than those of the copying mechanism; in addition, the qualitative-evaluation radiative performance of the former is higher.

• Journal of the Institute of Convergence Signal Processing
• /
• v.9 no.2
• /
• pp.129-134
• /
• 2008

In this paper, we carry out the simulation for an optimal solution of one-dimensional nonlinear photonic crystal structure using Genetic algorithm, and show the proposed method to apply for photonic transistors. Unlike a conventional steepest descent method for an optimization, the proposed method based on Genetic Algorithm has advantages for finding out excellent solutions without any analytic forms, which can easily apply to other applications. Also, as several solutions around global minimum solution can be obtained, it is very good optimization tool to give us the patterns about the optimal structure of a photonic crystal transistor. To design an all-optical filter transistor, Neural network algorithm is firstly performed for an initial design and then Genetic Algorithm is finally used to get the optimal solution. From the simulation of one-dimensional photonic crystal transistor, 27dB of the switching On/Off ratio is obtained.

• Journal of the Institute of Convergence Signal Processing
• /
• v.10 no.1
• /
• pp.86-92
• /
• 2009

As increasing of using induction motors, the induction motors faults cause serious damage to the industry. Therefore to find out faults of induction motor is recognized as important problem awaiting solution. But to make matters worse, the faults of induction motors often progress through long time. It means that early diagnosis is very important. Many researches have been progressed and general method of diagnosis is using vibration sensor to diagnose fault of induction motor. However, although it is reliability technique, it demands high price and it is difficult to use. This paper presents an implementation of technique for fault diagnosis of induction motor using wavelet transform based stator current and it is composed with algorithm that decides whether fault existence or not using C++ based on windows software. The algorithm will be accomplished in real-time using current data acquisition board and PC automatically with Neural Network algorithm.

• Journal of the Institute of Convergence Signal Processing
• /
• v.10 no.1
• /
• pp.93-99
• /
• 2009

The fuzzy control, neural network and genetic algorithm(GA) are algorithms to make the intelligence of system more higher. In this paper, we optimized the fuzzy controller using a genetic algorithm for desire response. Also a compensated fuzzy controller has dual rules. One control rule used to decrease the overshoot and rise time occurring in transient response region and another fuzzy control rule use to decrease the steady state error and rapildy to converge at the convergence region. GA is necessary to optimal the exchange time of the two fuzzy control rule base. Fuzzy-GA controller have a process of reproduction, crossover and mutation and we experimented by hydraulic servo motor control system We could observe that compensated Fuzzy-GA controller have good control performance compare to the fuzzy control technique have two rule base table.

• The Journal of the Korea Contents Association
• /
• v.8 no.11
• /
• pp.85-93
• /
• 2008

For music score recognition, several approaches have been proposed including shape matching, statistical methods, neural network based methods and structural methods. In this paper, we deal with recognition for low resolution images which are captured by the digital camera of a mobile phone. Considerable distortions are included in these low resolution images, so when existing technology is used, many problems appear. First, captured images are not stable in the sense that they contain lots of distortions or non-uniform illumination changes. Therefore, notes or symbols in the music score are damaged and recognition process gets difficult. This paper presents recognition technology to overcome these problems. First, musical note to head, stick, tail part are separated. Then template matching on head part of musical note, and remainder part is applied. Experimental results show nearly 100% recognition rate for music scores with single musical notes.

• The Journal of the Korea Contents Association
• /
• v.11 no.3
• /
• pp.271-280
• /
• 2011

Heart Rate Variability (HRV) analysis is a convenient tool to assess Myocardial Ischemia (MI). The analysis methods of HRV can be divided into time domain and frequency domain analysis. This paper uses wavelet transform as frequency domain analysis in contrast to time domain analysis in short term HRV analysis. ST-T and normal episodes are collected from the European ST-T database and the MIT-BIH Normal Sinus Rhythm database, respectively. An episode can be divided into several segments, each of which is formed by 32 successive RR intervals. Eighteen HRV features are extracted from each segment by the time and frequency domain analysis. To diagnose MI, the Neural Network with Weighted Fuzzy Membership functions (NEWFM) is used with the extracted 18 features. The results show that the average accuracy from time and frequency domain features is 75.29% and 80.93%, respectively.

• The KSFM Journal of Fluid Machinery
• /
• v.13 no.5
• /
• pp.43-53
• /
• 2010

This study investigates a design optimization of a rotating two-pass rectangular cooling channel with staggered arrays of pin-fins. The radial basis neural network method is used as an optimization technique with Reynolds-averaged Navier-Stokes analysis of fluid flow and heat transfer with shear stress transport turbulent model. The ratio of the diameter to height of the pin-fins and the ratio of the streamwise spacing between the pin-fins to height of the pin-fin are selected as design variables. The optimization problem has been defined as a minimization of the objective function, which is defined as a linear combination of heat transfer related term and friction loss related term with a weighting factor. Results are presented for streamlines, velocity vector fields, and contours of Nusselt numbers, friction coefficients, and turbulent kinetic energy. These results show how fluid flow in a two-pass square cooling channel evolves a converted secondary flows due to Coriolis force, staggered arrays of pin-fins, and a $180^{\circ}$ turn region. These results describe how the fluid flow affects surface heat transfer. The Coriolis force induces heat transfer discrepancy between leading and trailing surfaces, having higher Nusselt number on the leading surface in the second pass while having lower Nusselt number on the trailing surface. Dean vortices generated in $180^{\circ}$ turn region augment heat transfer in the turning region and in the upstream region of the second pass. As the result of optimization, in comparison with the reference geometry, thermal performance of the optimum geometry shows the improvement by 30.5%. Through the optimization, the diameter of pin-fin increased by 14.9% and the streamwise distance between pin-fins increased by 32.1%. And, the value of objective function decreased by 18.1%.

• Journal of Digital Convergence
• /
• v.18 no.2
• /
• pp.261-268
• /
• 2020

Machine learning is achieved by making a cost function from a given digital signal data and optimizing the cost function. The cost function here has local minimums in the cost function depending on the amount of digital signal data and the structure of the neural network. These local minimums make a problem that prevents learning. Among the many ways of solving these methods, our proposed method is to change the learning step-size. Unlike existed methods using the learning rate (step-size) as a fixed constant, the use of multivariate function as the cost function prevent unnecessary machine learning and find the best way to the minimum value. Numerical experiments show that the results of the proposed method improve about 3%(88.8%→91.5%) performance using the proposed method rather than the existed methods.

• Journal of the Korean Chemical Society
• /
• v.45 no.5
• /
• pp.422-428
• /
• 2001

Retention relative times(RRTs) of PAH molecules and their derivatives in gas chromatography are trained and predicted in testing sets using a multiple linear regression(MLR) and an artificial neural network(ANN). The main descriptors of PAHs and their derivatives in QSRR are the square root of molecular weight(sqmw), molecular connectivity($^1{\chi}_v$), molecular dipole moment(D) and length-to-breadth ratios(L/B). The results of MLR shows that a heavy molecule has a propensity for long retention time. L/B closely related with slot model is a good descriptor in MLR. On the other hand, ANN which is not effected by the linear dependencies among the descriptors were exclusively based on molecular weight and molecular dipole moment. The variances which shows the accuracy of prediction for retention times in testing sets are 1.860, 0.206 for MLR and ANN, respectively. It was shown that ANN can exceed the MLR in prediction accuracy.

• Proceedings of the KSRS Conference
• /
• 2007.10a
• /
• pp.438-441
• /
• 2007

Recently, due to molecular biology and engineering technology, DNA microarray makes people watch thousands of genes and the state of variation from the tissue samples of living body. With DNA Microarray, it is possible to construct a genetic group that has similar expression patterns and grasp the progress and variation of gene. This paper practices Cluster Analysis which purposes the discovery of biological subgroup or class by using gene expression information. Hence, the purpose of this paper is to predict a new class which is unknown, open leukaemia data are used for the experiment, and MCL (Markov CLustering) algorithm is applied as an analysis method. The MCL algorithm is based on probability and graph flow theory. MCL simulates random walks on a graph using Markov matrices to determine the transition probabilities among nodes of the graph. If you look at closely to the method, first, MCL algorithm should be applied after getting the distance by using Euclidean distance, then inflation and diagonal factors which are tuning modulus should be tuned, and finally the threshold using the average of each column should be gotten to distinguish one class from another class. Our method has improved the accuracy through using the threshold, namely the average of each column. Our experimental result shows about 70% of accuracy in average compared to the class that is known before. Also, for the comparison evaluation to other algorithm, the proposed method compared to and analyzed SOM (Self-Organizing Map) clustering algorithm which is divided into neural network and hierarchical clustering. The method shows the better result when compared to hierarchical clustering. In further study, it should be studied whether there will be a similar result when the parameter of inflation gotten from our experiment is applied to other gene expression data. We are also trying to make a systematic method to improve the accuracy by regulating the factors mentioned above.