• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.06a
• /
• pp.84-85
• /
• 2019

MPEG-NNR (Compression of Neural Network for Multimedia Content Description and Analysis) aims to define a compressed and interoperable representation of trained neural networks. In this paper, we present a low-rank approximation to compress a CNN used for image compression, which is one of MPEG-NNR use cases. In the presented method, the low-rank approximation decomposes one 2D kernel matrix of weights into two 1D kernel matrix values in each convolution layer to reduce the data amount of weights. The evaluation results show that the model size of the original CNN is reduced to half as well as the inference runtime is reduced up to about 30% with negligible loss in PSNR.

• Proceedings of the Korean Operations and Management Science Society Conference
• /
• 1996.04a
• /
• pp.100-103
• /
• 1996

Presented in this study is an algorithmic procedure to obtain polyhedral model from a compound surface. The compound surface in this study denotes a collection of trimmed surfaces without topological relations. The procedure consists of two main modules: CAD data interface, and surface conversion to polyhedral model. The interface module gets geometric information from CAD databases, and makes topological information by scanning the geometric information. We are investigating CATIA system as a data source system. In the surface conversion module, a shell(compound surface with topological information) is approximated to a triangular-faceted polyhedral surface model through node sampling and triangulation steps. The obtained polyhedral model should obey the vertex-to-vertex rule and meet tolerance requirements. Since the polyhedral model has a simple data structure and geometry processing for it is very efficient and robust, the polyhedral model can be used in various applications, such as surface rendering in computer graphics, FEM model for engineering analysis, CAPP for surface machining, data generation for SLA, and NC tool path generation.

• Journal of the Korean Institute of Illuminating and Electrical Installation Engineers
• /
• v.17 no.5
• /
• pp.111-117
• /
• 2003

The Taylor model concept is introduced to design a controller with input and output data only. The parameters in Taylor model can be estimated using the input and output data and a controller can be designed based on Taylor model. The accuracy of Taylor model approximation can be improved by increasing the observation window and the order of Taylor model. The LQR method is applied to Taylor model to design power system stabilizers (PSS), and compared with the conventional PSS.

• Journal of the Korean Statistical Society
• /
• v.25 no.3
• /
• pp.433-444
• /
• 1996

In contrast to the multiplicative risk model, the additive risk model specifies that the hazard function with covariates is the sum of, rather than product of, the baseline hazard function and the regression function of covariates. We, in this paper, propose a method for checking the adequacy of the additive risk model based on partial-sum of matingale residuals. Under the assumed model, the asymptotic properties of the proposed test statistic and approximation method to find the critical values of the limiting distribution are studied. Several real examples are illustrated.

• Proceedings of the KSRS Conference
• /
• 2003.11a
• /
• pp.750-752
• /
• 2003

The objective of this investigation is to compare the geometric precision of Rigorous Sensor Model and Rational Function Model for QuickBird images. In rigorous sensor model, we use the on-board data and ground control points to fit an orbit; then, a least squares filtering technique is applied to collocate the orbit. In rational function model, we first use the rational polynomial coefficients provided by the satellite company. Then the systematic bias of the coefficients is compensated by an affine transformation using ground control points. Experimental results indicate that, the RFM provides a good approximation in the position accuracy.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.56 no.7
• /
• pp.1315-1320
• /
• 2007

In this paper, we introduce a new fuzzy model called fuzzy combined polynomial neural networks, which are based on the representative fuzzy model named polynomial fuzzy model. In the design procedure of the proposed fuzzy model, the coefficients on consequent parts are estimated by using not general least square estimation algorithm that is a sort of global learning algorithm but weighted least square estimation algorithm, a sort of local learning algorithm. We are able to adopt various type of structures as the consequent part of fuzzy model when using a local learning algorithm. Among various structures, we select Polynomial Neural Networks which have nonlinear characteristic and the final result of which is a complex mathematical polynomial. The approximation ability of the proposed model can be improved using Polynomial Neural Networks as the consequent part.

• Journal of the Korean Institute of Electrical and Electronic Material Engineers
• /
• v.21 no.1
• /
• pp.73-78
• /
• 2008

A resistor network model for the horizontal AlInGaN LED was investigated, The parameters of the proposed model are extracted from the test dies and $350{\mu}m$ LED, The center of the P-area is the optimal position of a P-electrode by the simulation using the model. Also the optimal chip size of the LED for the new target current was investigated, Comparing the simulation and fabrication result, the errors for the forward voltage and the light power are average 0,02 V, 8 % respectively, So the proposed resistor network model with the linear forward voltage approximation and the exponential light power model are useful in the simulation for the horizontal AlInGaN LED.

• JSTS:Journal of Semiconductor Technology and Science
• /
• v.14 no.4
• /
• pp.451-456
• /
• 2014

A compact model of a depletion-mode silicon-nanowire (Si-NW) pH sensor is proposed. This drain current model is obtained from the Pao-Sah integral and the continuous charge-based model, which is derived by applying the parabolic potential approximation to the Poisson's equation in the cylindrical coordinate system. The threshold-voltage shift in the drain-current model is obtained by solving the nonlinear Poisson-Boltzmann equation for the electrolyte. The simulation results obtained from the proposed drain-current model for the Si-NW field-effect transistor (SiNWFET) agree well with those of the three-dimensional (3D) device simulation, and those from the Si-NW pH sensor model also agree with the experimental data.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.14 no.3
• /
• pp.619-633
• /
• 1994

A wave deformation model for general purpose combined refraction, diffraction, and breaking is developed in the shallow water. A parabolic approximation equation considered a higher order diffraction term is derived from the previous mild slope equation. A wave energy dissipation term due to bottom friction and breaking is introduced from the turbulence model. The Crank-Nicoloson implicit scheme is used in the numerical calculation, then the solutions are compared with the various hydraulic experiment data in the circular, the elliptic shoal, and the surf zone. The wave height decay in the surf zone is sensitively affected by the incident wave steepness, and the wave height variation around the elliptic shoal is well explained by the non-linear dispersion relation and the wave energy dissipation term. The model is also applied to a field coastal area and reasonable results are obtained.

• Journal of the Korean Institute of Telematics and Electronics D
• /
• v.36D no.7
• /
• pp.9-16
• /
• 1999

We have investigated effects of local damage accumulation for ultra-low energy As and B ion implant using highly efficient molecular dynamics(MD) scheme. We simulated ion implantation by MD simulation using recoil ion approximation (RIA) method and local cell damage accumulation (LCDA) model proposed in the paper. Local damage accumulation probability function consisted of deposited energy in a unit cell, implant dose rate, target material, projectile atom, and recoil event number. The simulated results were good agreement with the experimental and other simulated results. The MDRANGE results without damage accumulation were different from SIMS data in the tail region. We also simulated 2 dimensional dopant and damage profiles using the local damage accumulation model and recoil ion approximation method.