• Journal of Advanced Marine Engineering and Technology
• /
• v.21 no.1
• /
• pp.66-70
• /
• 1997

In this paper a new learning algorithm for curvature smoothing and improved generalization for multi-layer neural networks is proposed. To enhance the generalization ability a constraint term of hidden neuron activations is added to the conventional output error, which gives the curvature smoothing characteristics to multi-layer neural networks. When the total cost consisted of the output error and hidden error is minimized by gradient-descent methods, the additional descent term gives not only the Hebbian learning but also the synaptic weight decay. Therefore it incorporates error back-propagation, Hebbian, and weight decay, and additional computational requirements to the standard error back-propagation is negligible. From the computer simulation of the time series prediction with Santafe competition data it is shown that the proposed learning algorithm gives much better generalization performance.

• Proceedings of the Korean Society for Noise and Vibration Engineering Conference
• /
• 2007.05a
• /
• pp.1023-1026
• /
• 2007

Optimal damping layout of the constrained viscoelastic damping layer on beam is identified with temperatures by using a gradient-based numerical search algorithm. An optimal design problem is defined in order to determine the constrained damping layer configuration. A finite element formulation is introduced to model the constrained damping layer beam. The four-parameter fractional derivative model and the Arrhenius shift factor are used to describe dynamic characteristics of viscoelastic material with respect to frequency and temperature. Frequency-dependent complex-valued eigenvalue problems are solved by using a simple resubstitution algorithm in order to obtain the loss factor of each mode and responses of the structure. The results of the numerical example show that the proposed method can reduce frequency responses of beam at peaks only by reconfiguring the layout of constrained damping layer within a limited weight constraint.

• Atmosphere
• /
• v.15 no.3
• /
• pp.155-165
• /
• 2005

The characteristics of six precipitation systems occurred around Cheongju in 2002 are analyzed after the convective/stratiform radar echo classification using radar reflectivity from the Meteorological Research Institute"s X-band Doppler weather radar. The Biggerstaff and Listemaa (2000) algorithm is applied for the classification and reveals a physical characteristics of the convective and stratiform rain diagnosed from the three-dimensional structure of the radar reflectivity. The area satisfying the vertical profile of radar reflectivity is well classified, while the area near the radar site and the topography-shielded area show a mis-classification. The seasonal characteristics of the precipitation system are also analyzed using the contoured frequency by altitude diagrams (CFADs). The heights of maximum reflectivity are 4 km and 5.5 km in spring and summer, respectively, and the vertical gradient of radar reflectivity from 1.5 km to the melting layer in spring is larger than in summer.

• Proceedings of the KIEE Conference
• /
• 2006.07d
• /
• pp.1858-1859
• /
• 2006

This paper presents a new fuzzy-neural-network based interacting multiple model (FNNBIMM) algorithm for tracking a maneuvering target. To effectively handle the unknown target acceleration, this paper regards it as additional noise, time-varying variance to target model. Each sub model characterized by the variance of the overall process noise, which is obtained on the basis of each acceleration interval. Since it is hard to approximate this time-varying variance adaptively owing to the unknown acceleration, the FNN is utilized to precisely approximate this time-varying variance. The gradient descendant method is utilized to optimize each FNN. To show the feasibility of the proposed algorithm, a numerical example is provided.

• Journal of the Korea Society for Simulation
• /
• v.5 no.2
• /
• pp.85-93
• /
• 1996

Simulation is commonly used to find the best values of decision variables for problems which defy analytical solutions. This objective is similar to that of optimization problems and thus, mathematical programming techniques may be applied to simulation. However, the application of mathematical programming techniques, e.g., the gradient methods, to simulation is compounded by the random nature of simulation responses and by the complexity of the statistical issues involved. In this paper, therefore, we explain the Reverse-Simulation method to optimize a simulation model in a single simulation run. First, we point the problem of the previous Reverse-Simulation method. Secondly, we propose the new algorithm to solve the previous method and show the efficiency of the proposed algorithm.

• Proceedings of the IEEK Conference
• /
• 1998.06a
• /
• pp.661-664
• /
• 1998

Since the region segmentation at high resolution contains most of viable semantic object contours in an image, the bottom-up approach for image segmentation is appropriate for the application such as MPEG-4 which needs to preserve semantic object contours. However, the conventioal region merging methods, that follow the region segmentation, have poor performance in keeping low-contrast semantic object contours. In this paper, we propose an image segmentation algorithm based on classified region merging. The algorithm pre-segments an image with a large number of small regions, and also classifies it into several classes having similar gradient characteristics. Then regions only in the same class are merged according to the boundary weakness or statisticsal similarity. The simulation result shows that the proposed image segmentation preserves semantic object contours very well even with a small number of regions.

• Proceedings of the Korean Society of Precision Engineering Conference
• /
• 2003.06a
• /
• pp.1362-1365
• /
• 2003

On this study. we improved the efficiency applying algorithm that is repeatedly using orthogonal array in discrete design space and filling a defect of gradient method in continuous design space. we showed optimal ply angle that maximized buckling strength of CFRP laminated composite plate without a hole and with a hole by each aspect ratio. In the case of CFRP laminated composite plate without a hole, we confirmed the reliance and efficiency of algorithm in comparison with the result optimization achievement repeatedly using statistical orthogonal array of experimental design.

• Structural Engineering and Mechanics
• /
• v.21 no.2
• /
• pp.121-136
• /
• 2005

This paper presents the implementation of the Blatz-Ko hyperelastic compressible model in a finite element program to deal with large deformation problems. We show analytically and numerically that the minimum number of increment steps in the Newton-Raphson algorithm depends on material properties and applied loads. We also show that this dependence is related to the orientation preservation principle. So we propose a convergence criteria based on the sign of eigenvalues of the deformation gradient matrix.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2004.10a
• /
• pp.63-66
• /
• 2004

The SGA(stochastic gradient ascent) algorithm is one of the most important tools in the area of reinforcement learning, and has been applied to a wide range of practical problems. In particular, this learning method was successfully applied by Kimura et a1. [1] to the control of a simple creeping robot which has finite number of control input choices. In this paper, we considered the application of the SGA algorithm to Kimura's robot control problem for the case that the control input is not confined to a finite set but can be chosen from a infinite subset of the real numbers. We also developed a MATLAB-based robot animation program, which showed the effectiveness of the training algorithms vividly.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.29 no.5C
• /
• pp.700-705
• /
• 2004

Clustering of incomplete data using the Autoencoder and the Fuzzy c-Means(PCM) is proposed in this paper. The Proposed algorithm, called Optimal Completion Autoencoder Fuzzy c-Means(OCAEFCM), utilizes the Autoencoder Neural Network (AENN) and the Gradiant-based FCM (GBFCM) for optimal completion of missing data and clustering of the reconstructed data. The proposed OCAEFCM is applied to the IRIS data and a data set from a financial institution to evaluate the performance. When compared with the existing Optimal Completion Strategy FCM (OCSFCM), the OCAEFCM shows 18％-20％ improvement of performance over OCSFCM.