• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.32B no.11
• /
• pp.1489-1495
• /
• 1995

In this paper, we propose recurrent neural prediction models (RNPM), recurrent neural networks trained as a nonlinear predictor of speech, as a new connectionist model for speech recognition. RNPM modulates its mapping effectively by internal representation, and it requires no time alignment algorithm. Therefore, computational load at the recognition stage is reduced substantially compared with the well known predictive neural networks (PNN), and the size of the required memory is much smaller. And, RNPM does not suffer from the problem of deciding the time varying target function. In the speaker dependent and independent speech recognition experiments under the various conditions, the proposed model was comparable in recognition performance to the PNN, while retaining the above merits that PNN doesn't have.

• Journal of the Korean Institute of Telematics and Electronics C
• /
• v.34C no.1
• /
• pp.42-50
• /
• 1997

Vector quantization has shown to be useful for compressing data related with a wide rnage of applications such as image processing, speech processing, and weather satellite. Neural networks of images this paper propses a efficient neural network learning algorithm, called learning count control algorithm based on the frquency sensitive learning algorithm. This algorithm can train a results more codewords can be assigned to the sensitive region of the human visual system and the quality of the reconstructed imate can be improved. We use a human visual systrem model that is a cascade of a nonlinear intensity mapping function and a modulation transfer function with a bandpass characteristic.

• Proceedings of the Computational Structural Engineering Institute Conference
• /
• 1997.04a
• /
• pp.53-60
• /
• 1997

The strain localization of concrete is a phenomenon such that the deformation of concrete is localized in finite region along with softening behavior. The objective of this paper is to develope a consistent algorithm for the finite element modeling of localized zone in the analysis of the strain-localization in concrete. For modeling of the localized zone in concrete under strain localization, a general Drucker-Prager failure criterion which can consider nonlinear strain softening behavior of concrete after peak-stress is introduce. The return-mapping algorithm is used for the integration of the elasto-plastic rate equation and the consistent tangent modulus is derived. Using finite element program implemented with the developed algorithms, strain localization behaviors for the different sizes of concrete specimen under compression are simulated.

• Journal of Electrical Engineering and information Science
• /
• v.2 no.5
• /
• pp.51-59
• /
• 1997

This paper presets a methodology of improving a conventional model in power systems using Genetic Algorithms(GAs) and suggests a GAs-based model which can directly solve the real-valued optimum in an optimization procedure. In applying GAs to the power flow, a new fitness mapping method is proposed using the proposed using the probability distribution function for all the payoffs in the population pool. In this approach, both the notions on a way of the genetic representation, and a realization of the genetic operators are fully discussed to evaluate he GAs' effectiveness. The proposed method is applied to IEEE 5-bus, 14-bus and 25-bus systems and, the results of computational experiments suggest a direct applicability of GAs to more complicated power system problems even if they contain nonlinear algebraic equations.

• The Transactions of the Korea Information Processing Society
• /
• v.6 no.10
• /
• pp.2716-2721
• /
• 1999

The problem of determining the conformal maps from the unit disk onto a jordan region has been completed by solving the theodorsen equation that is nonlinear. For the hubners method, which has been well known for the efficient method among the many suggestions for the Theodorsen equation, it has been reproved in our early study that the convergence rate could be remarkably improved by exploring and applying a low-frequency pass filter[1]. However, in the Hubner's method with the low-frequency filter, the discrete numbers and parameters of the low-frequency filter were able to be acquired only by experience. In this paper we show algorithms that determine the discrete numbers and parameters of the low-frequency filter automatically in accordance with the given region. This results from analyzing the function, which decides the shape of the given domain under the assumption that the degree of the problem depends of the transformation of a given domain, as seen in the Fourier Transform.

• Proceedings of the KIEE Conference
• /
• 2003.07d
• /
• pp.2289-2291
• /
• 2003

NLPCA(Nonlinear Principal Component Analysis is a novel technique for multivariate data analysis, similar to the well-known method of principal component analysis. NLPCA can be implemented by a feedforward neural network called AANN (AutoAssociative Neural Network) which performs the identity mapping. In this work, a sensor fault detection system based on NLPCA and Maximum Likelihood Estimation scheme is presented. To verify its applicability, simulation study on the data supplied from Saemangeum measurement stations is executed.

• Journal of Sensor Science and Technology
• /
• v.14 no.2
• /
• pp.109-115
• /
• 2005

Calibration is a prerequisite procedure for employing a camera as a 3D sensor in an automated machines like robots. As accurate sensing is possible only when the vision sensor is calibrated accurately, many different approaches and models have been proposed for increasing calibration accuracy. Particularly an important factor which greatly affects the calibration accuracy is the nonlinearity in the mapping between 3D world and corresponding 2D image. In this paper GMDH algorithm is used to learn the nonlinearity without physical modelling. The technique proposed can be effective in various situations where the levels of noises and characteristics of nonlinear distortion are different. In simulations and an experiment, the proposed technique showed good and reliable results.

• Water for future
• /
• v.29 no.4
• /
• pp.109-118
• /
• 1996

The neural network is a mathematical model of theorized brain activity which attempts to exploit the parallel local processing and distributed storage properties. The neural metwork is a good model to be applied for the classification problem, large combinatorial optimization and nonlinear mapping. A multi-layer neural network is constructed to predict rainfall. The network learns continuourvalued input and output data. Application of neural network to 1-hour real data in Seoul metropolitan area and the Soyang River basin shows slightly good predictions. Therefore, when good data is available, the neural network is expected to predict the complicated rainfall successfully.

• Bulletin of the Korean Mathematical Society
• /
• v.37 no.1
• /
• pp.153-169
• /
• 2000

Let E be a real q-uniformly smooth Banach space which admits a weakly sequentially continuous duality map, and K a nonempty closed convex subset of E. Let T : K -> K be a mapping such that $F(T)\;=\;{x\;{\in}\;K\;:\;Tx\;=\;x}\;{\neq}\;0$ and (I - T) satisfies the accretive-type condition: $\;{\geq}\;{\lambda}$\mid$$\mid$x-Tx$\mid$$\mid$^2$, for all $x\;{\in}\;K,\;x^*\;{\in}\;F(T)$ and for some ${\lambda}\;>\;0$. The weak and strong convergence of the Ishikawa iteration method to a fixed point of T are investigated. An application of our results to the approximation of a solution of a certain linear operator equation is also given. Our results extend several important known results from the Mann iteration method to the Ishikawa iteration method. In particular, our results resolve in the affirmative an open problem posed by Naimpally and Singh (J. Math. Anal. Appl. 96 (1983), 437-446).

• Structural Engineering and Mechanics
• /
• v.10 no.2
• /
• pp.165-180
• /
• 2000

An analytical model with tension softening for the prediction of the capacity of prestressed concrete beams under pure torsion and under torsion combined with shear and flexure is introduced. The proposed approach employs bilinear stress-strain relationship with post cracking tension softening branch for the concrete in tension and special failure criteria for biaxial stress states. Further, for the solution of the governing equations a special numerical scheme is adopted which can be applied to elements with practically any cross-section since it utilizes a numerical mapping. The proposed method is mainly applied to plain prestressed concrete elements, but is also applicable to prestressed concrete beams with light transverse reinforcement. The aim of the present work is twofold; first, the validation of the approach by comparison between experimental results and analytical predictions and second, a parametrical study of the influence of concentric and eccentric prestressing on the torsional capacity of concrete elements and the interaction between torsion and shear for various levels of prestressing. The results of this investigation presented in the form of interaction curves, are compared to experimental results and code provisions.