• Journal of Broadcast Engineering
• /
• v.26 no.2
• /
• pp.125-131
• /
• 2021

In recent years, Convolutional Neural Networks (CNNs) have achieved outstanding performance in the fields of computer vision such as image classification, object detection, visual quality enhancement, etc. However, as huge amount of computation and memory are required in CNN models, there is a limitation in the application of CNN to low-power environments such as mobile or IoT devices. Therefore, the need for neural network compression to reduce the model size while keeping the task performance as much as possible has been emerging. In this paper, we propose a method to compress CNN models by combining matrix decomposition methods of LR (Low-Rank) approximation and CP (Canonical Polyadic) decomposition. Unlike conventional methods that apply one matrix decomposition method to CNN models, we selectively apply two decomposition methods depending on the layer types of CNN to enhance the compression performance. To evaluate the performance of the proposed method, we use the models for image classification such as VGG-16, RestNet50 and MobileNetV2 models. The experimental results show that the proposed method gives improved classification performance at the same range of 1.5 to 12.1 times compression ratio than the existing method that applies only the LR approximation.

• Journal of the Earthquake Engineering Society of Korea
• /
• v.11 no.6
• /
• pp.53-61
• /
• 2007

This study investigates the effect of site class, post-yield stiffness ratio, damping ratio, yield-strength reduction factor, and natural period on inelastic displacement ratio of bilinear SDF systems located at the sites classified as NEHRP site class B,C,D. The previous studies developed inelastic displacement ratio using equal displacement rule in the intermediate and long period range. But, this approximation overestimates the inelastic displacement ratio. Furthermore, inelastic displacement ratio has not been developed for the systems having a damping ratio less than 5%. This study conducts nonlinear regression analysis for proposing equations for calculating median and deviation of the inelastic displacement ratio of the bilinear SDOF system having damping ratios ranging from 0 to 20%. Using median and deviation of the inelastic displacement ratio, probabilistic inelastic displacement ratio is estimated, which can be used for performance-based seismic evaluation.

• Tunnel and Underground Space
• /
• v.13 no.5
• /
• pp.389-396
• /
• 2003

It is shown that the cubic law can be modified regarding the steady-state Navier-Stokes equations by using perturbation approximation method for a sinusoidal aperture variation. In order to adopt the perturbation theory, the sinusoidal function needs to be non-dimensionalized for the amplitude and wavelength. Then, the steady-state Navier-Stokes equations can be solved by expanding the non-dimensionalized stream function with respect to the small value of the parameter (the ratio of the mean aperture to the wavelength), together with the continuity equation. From the approximate solution of the Navier-Stokes equations, the basic cubic law is successfully modified for the steady-state condition and a sinusoidal aperture variation. A finite difference method is adopted to calculate the pressure within a fracture model, and the results of numerical experiments show the accuracy and applicability of the modified cubic law. As a result, it is noted that the modified cubic law, suggested in this study, will be used for the analysis of fluid flow through aperture geometry of sinusoidal distributions.

• Journal of Internet Computing and Services
• /
• v.12 no.5
• /
• pp.11-19
• /
• 2011

In this paper, Service Boundary Line approximation method is proposed to improve the accuracy of aggregated link state information for source routing in transport networks that conduct hierarchical QoS routing. The proposed method is especially useful for aggregating links that have both delay and bandwidth as their QoS parameters. This method selects the main path weight in the network and transports the data to the external networks together with the aggregation information, reducing information distortion caused from the loss of some path weight during aggregation process. In this paper, the main path weight is defined as outlier. Service Boundary Line has 2k+5parameters. k is the number of outliers. The number of storage spaces of Service Boundary Line changes according to the number of outliers. Simulation results show that our approximation method requires a storage space that 1.5-2 times larger than those in other known techniques depending on outlier selection method, but its information accuracy of proposed method in the ratio between storage space and information accuracy is higher.

• Investigative Magnetic Resonance Imaging
• /
• v.20 no.4
• /
• pp.215-223
• /
• 2016

Purpose: The management of metal-induced field inhomogeneities is one of the major concerns of distortion-free magnetic resonance images near metallic implants. The recently proposed method called "Slice Encoding for Metal Artifact Correction (SEMAC)" is an effective spin echo pulse sequence of magnetic resonance imaging (MRI) near metallic implants. However, as SEMAC uses the noisy resolved data elements, SEMAC images can have a major problem for improving the signal-to-noise ratio (SNR) without compromising the correction of metal artifacts. To address that issue, this paper presents a novel reconstruction technique for providing an improvement of the SNR in SEMAC images without sacrificing the correction of metal artifacts. Materials and Methods: Low-rank approximation in each coil image is first performed to suppress the noise in the slice direction, because the signal is highly correlated between SEMAC-encoded slices. Secondly, SEMAC images are reconstructed by the best linear unbiased estimator (BLUE), also known as Gauss-Markov or weighted least squares. Noise levels and correlation in the receiver channels are considered for the sake of SNR optimization. To this end, since distorted excitation profiles are sparse, $l_1$ minimization performs well in recovering the sparse distorted excitation profiles and the sparse modeling of our approach offers excellent correction of metal-induced distortions. Results: Three images reconstructed using SEMAC, SEMAC with the conventional two-step noise reduction, and the proposed image denoising for metal MRI exploiting sparsity and low rank approximation algorithm were compared. The proposed algorithm outperformed two methods and produced 119% SNR better than SEMAC and 89% SNR better than SEMAC with the conventional two-step noise reduction. Conclusion: We successfully demonstrated that the proposed, novel algorithm for SEMAC, if compared with conventional de-noising methods, substantially improves SNR and reduces artifacts.

• Journal of the Institute of Electronics Engineers of Korea TC
• /
• v.45 no.7
• /
• pp.9-16
• /
• 2008

For the efficient routing on a Sensor Network, one may consider a deployment problem to interconnect the sensor nodes optimally. There is an analogous theoretic problem: the Steiner Tree problem of finding the tree that interconnects given points on a plane optimally. One may use the approximation algorithm for the problem to find out the deployment that interconnects the sensor nodes almost optimally. However, the Steiner Tree problem is to interconnect mathematical set of points on a Euclidean plane, and so involves particular cases that do not occur on Sensor Networks. Thus the approach of using the algorithm does not make a proper way of analysis. Differently from the randomly given locations of mathematical points on a Euclidean plane, the locations of sensors on Sensor Networks are assumed to be physically dispersed over some moderate distance with each other. By designing an approximation algorithm for the Sensor Networks in terms of that physical property, one may produce the execution time and the approximation ratio to the optimality that are appropriate for the problem of interconnecting Sensor Networks.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.25 no.10A
• /
• pp.1566-1575
• /
• 2000

This paper studies mismatched scalar quantization of a generalized gamma source by a quantizer that is optimally (in the mean square error sense) designed for another generalized gamma source. Specifically, it considers variance-mismatched quantization which occurs when the variance of the source to be quantized differs from tat of the designed-for source. The main result is the two distortion formulas derived from Bennett's integral. The first formula is an approximation expression that uses the outermost threshold of an optimum scalar quantizer, and the second formula, in turn, uses an approximation formula for this outermost threshold. Numerical results are obtained for Laplacian sources, which are example of a generalized gamma source, and comparisons are made between actual mismatched distortions and the two formulas. These numerical results show that the two formulas become more accurate, as the number of quantization points gets larger and the ratio of the source variance to that of the designed-for source gets bigger. For example, the formulas are within 2~4% of the actual distortion for approximately 64 quantization points or more. In conclusion, the proposed approximation formulas are considered to have contribution as closed formulas and for their accuracy.

• The Korean Journal of Applied Statistics
• /
• v.31 no.3
• /
• pp.397-408
• /
• 2018

Various methods for accurate parameter estimation have been developed in a sample survey and it is also common to use a ratio estimator or the regression estimator using auxiliary information. The ratio-type estimator has been used in many recent studies and is known to improve the accuracy of estimation by adjusting the ratio estimator. However, various studies are under way to solve it since the ratio-type estimator is biased. In this study, we propose a generalized ratio-type estimator with a new parameter added to the ratio-type estimator to remove the bias. We suggested a method to apply this result to the parameter estimation under the error assumption of heteroscedasticity. Through simulation, we confirmed that the suggested generalized ratio-type estimator gives good results compared to conventional ratio-type estimators.

• International Journal of Fluid Machinery and Systems
• /
• v.3 no.2
• /
• pp.150-159
• /
• 2010

Laidback fan shaped film-cooling hole is formulated numerically and optimized with the help of three-dimensional numerical analysis, surrogate methods, and the multi-objective evolutionary algorithm. As Pareto optimal front produces a set of optimal solutions, the trends of objective functions with design variables are predicted by hybrid multi-objective evolutionary algorithm. The problem is defined by four geometric design variables, the injection angle of the hole, the lateral expansion angle of the diffuser, the forward expansion angle of the hole, and the ratio of the length to the diameter of the hole, to maximize the film-cooling effectiveness compromising with the aerodynamic loss. The objective function values are numerically evaluated through Reynolds- averaged Navier-Stokes analysis at the designs that are selected through the Latin hypercube sampling method. Using these numerical simulation results, the Response Surface Approximation model are constructed for each objective function and a hybrid multi-objective evolutionary algorithm is applied to obtain the Pareto optimal front. The clustered points from Pareto optimal front were evaluated by flow analysis. These designs give enhanced objective function values in comparison with the experimental designs.

• Journal of electromagnetic engineering and science
• /
• v.5 no.2
• /
• pp.49-60
• /
• 2005

In this paper, an efficient performance enhancement scheme for an adaptive antenna array under the flat and the frequency-selective Rayleigh fadings is proposed. The proposed signal enhancement scheme is the modified linear signal estimator which combines the rank N approximation by reducing noise eigenvalues(RANE) and Toeplitz matrix approximation(TMA) methods into the linear signal estimator. The proposed performance enhancement scheme is performed by not only reducing the noise component from the signal-plus-noise subspace using RANE but also having the theoretical property of noise-free signal using TMA. Consequently, the key idea of the proposed performance enhancement scheme is to greatly enhance the performance of an adaptive antenna array by removing all undesired noise effects from the post-correlation received signal. The proposed performance enhancement scheme applies at the Wiener maximal ratio combining(MRC) method which has been widely used as the conventional adaptive antenna array. It is shown through several simulation results that the performance of an adaptive antenna array using the proposed signal enhancement scheme is much superior to that of a system using the conventional method under several environments, i.e., a flat Rayleigh fading, a fast frequency-selective Rayleigh fading, a perfect/imperfect power control, a single cell, and a multicell.