• The Korean Journal of Applied Statistics
• /
• v.26 no.1
• /
• pp.71-79
• /
• 2013

Semiparametric and nonparametric small area estimations have been studied to overcome a large variance due to a small sample size allocated in a small area. In this study, we investigate semiparametric and nonparametric mixed effect small area estimators using penalized spline and kernel smoothing methods respectively and compare their performances using labor statistics.

• 한국가시화정보학회:학술대회논문집
• /
• 2004.12a
• /
• pp.47-55
• /
• 2004

Several interpolations for image deformation in PIV were evaluated. The tested interpolation methods are linear, quadratic, truncated sinc, windowed sinc, cubic, Lagrange, Gaussian $2^{nd}\;and\;6^{th}$ interpolators. Bias errors and random errors were evaluated in the range of $0\~3.0$ pixel uniform displacement using synthetic images. We also measured the time cost of each interpolator with respect to kernel size. The cubic interpolator with $6\times6$ kernel showed the best results in terms of the performance and time cost.

• ETRI Journal
• /
• v.39 no.6
• /
• pp.841-850
• /
• 2017

We investigate neural network image reconstruction for magnetic particle imaging. The network performance strongly depends on the convolution effects of the spectrum input data. The larger convolution effect appearing at a relatively smaller nanoparticle size obstructs the network training. The trained single-layer network reveals the weighting matrix consisting of a basis vector in the form of Chebyshev polynomials of the second kind. The weighting matrix corresponds to an inverse system matrix, where an incoherency of basis vectors due to low convolution effects, as well as a nonlinear activation function, plays a key role in retrieving the matrix elements. Test images are well reconstructed through trained networks having an inverse kernel matrix. We also confirm that a multi-layer network with one hidden layer improves the performance. Based on the results, a neural network architecture overcoming the low incoherence of the inverse kernel through the classification property is expected to become a better tool for image reconstruction.

• 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.

• Journal of KIISE:Computer Systems and Theory
• /
• v.29 no.3
• /
• pp.143-152
• /
• 2002

As the multimedia application becomes ubiquitous, the size of message used for Inter Process Communication (IPC) grows up to cope with the requirements of multimedia applications. And the rapid development of new hardware platforms makes the portability of operating system more important. But the traditional micro-kernel operating system is Implemented platform dependently for better performance, and especially focused on handling short message. In this paper, we present the design and implementation of IPC mechanism in M3K (MultiMedia Micro-Kernel) to address the above problems. Our IPC mechanism provides enhanced performance and efficiently handles large message without performance degrading.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.23 no.5
• /
• pp.931-938
• /
• 2013

It is hard to extract original data from encrypted data before getting the password in encrypted data with disk encryption software. This encryption key of disk encryption software can be extract by using physical memory analysis. Searching encryption key time in the physical memory increases with the size of memory because it is intended for whole memory. But physical memory data includes a lot of data that is unrelated to encryption keys like system kernel objects and file data. Therefore, it needs the method that extracts valid data for searching keys by analysis. We provide a method that collect only saved memory parts of disk encrypting keys in physical memory by analyzing Windows kernel virtual address space. We demonstrate superiority because the suggested method experimentally reduces more of the encryption key searching space than the existing method.

• Journal of KIISE:Computing Practices and Letters
• /
• v.6 no.2
• /
• pp.173-186
• /
• 2000

Real-time embedded systems should accommodate many kinds of hardware platforms and resource management policies that vary depending on their operating environments and purposes. It is not an easy job to adapt a multitasking kernel to new system services and hardware platforms, as the kernel must strictly satisfy constraints on its size and performance. This paper describes the design and implementation of an object-oriented multitasking framework that can be reused for implementing microprocessor-based real-time embedded multitasking kernels, In this kernel framework, those parts that can vary depending on hardware platforms and system resource management policies are separated into the hot spots and encapsulated by abstract classes. Our framework thus can be effectively used to implement microprocessor-based real-time embedded kernels that demand high portability and adaptability.

• Interaction and multiscale mechanics
• /
• v.2 no.1
• /
• pp.15-30
• /
• 2009

The theoretical framework of the interaction fields for multiple scales based on field theory is applied to one-dimensional problem mimicking dislocation substructure sensitive intra-granular inhomogeneity evolution under fatigue of Cu-added steels. Three distinct scale levels corresponding respectively to the orders of (A)dislocation substructures, (B)grain size and (C)grain aggregates are set-up based on FE-RKPM (reproducing kernel particle method) based interpolated strain distribution to obtain the incompatibility term in the interaction field. Comparisons between analytical conditions with and without the interaction, and that among different cell size in the scale A are simulated. The effect of interaction field on the B-scale field evolution is extensively examined. Finer and larger fluctuation is demonstrated to be obtained by taking account of the field interactions. Finer cell size exhibits larger field fluctuation whereas the coarse cell size yields negligible interaction effects.

• KOREAN JOURNAL OF CROP SCIENCE
• /
• v.61 no.3
• /
• pp.196-207
• /
• 2016

In monsoon climates, persistent shade is a troublesome weather condition with an impact on the growth and yield of corn (Zea mays L.). We imposed 7, 14, 21, and 28 days of consecutive shade (CS) on Gwangpyeongok and P3394 corn hybrids at the beginning of the kernel filling stages. Shade had little impact on leaf area and dry matter accumulation in the stem and leaves. However, dry matter accumulation in the ear was severely reduced by approximately 28% and 53% after 14 and 28 days of CS, respectively. For the components of grain yield, 7 and 14 days of shade did irreparable damage to the number of filled kernels, the kernel number per ear row, and the percent of filled kernels, but did little damage or reversible damage after removal of the shade to the 100-grain weight and the row number per ear. Shade significantly reduced the relative growth rate (RGR) due to a decrease in the net assimilation rate (NAR). These results suggest that source activity limitation by shade during the kernel filling stages leads to the inhibition of sink activity and size. The yield of biomass, ear, and grain logistically declined as the length of CS increased. Probit analysis revealed that the number of days of CS needed to cause 25% and 50% reductions in grain yield were 3.7 and 23.1, respectively. These results suggest that the plant yield loss induced by shade at the beginning of the kernel filling stages is mainly achieved within the first 7 days of consecutive shade.

• Investigative Magnetic Resonance Imaging
• /
• v.17 no.4
• /
• pp.275-285
• /
• 2013

Purpose : The objective of this study was to investigate effects of different smoothing kernel sizes on brain tissue-masked susceptibility-weighted images (SWI) obtained from normal elderly subjects using voxel-based analyses. Materials and Methods: Twenty healthy human volunteers (mean $age{\pm}SD$ = $67.8{\pm}6.09$ years, 14 females and 6 males) were studied after informed consent. A fully first-order flow-compensated three-dimensional (3D) gradient-echo sequence ran to obtain axial magnitude and phase images to generate SWI data. In addition, sagittal 3D T1-weighted images were acquired with the magnetization-prepared rapid acquisition of gradient-echo sequence for brain tissue segmentation and imaging registration. Both paramagnetically (PSWI) and diamagnetically (NSWI) phase-masked SWI data were obtained with masking out non-brain tissues. Finally, both tissue-masked PSWI and NSWI data were smoothed using different smoothing kernel sizes that were isotropic 0, 2, 4, and 8 mm Gaussian kernels. The voxel-based comparisons were performed using a paired t-test between PSWI and NSWI for each smoothing kernel size. Results: The significance of comparisons increased with increasing smoothing kernel sizes. Signals from NSWI were greater than those from PSWI. The smoothing kernel size of four was optimal to use voxel-based comparisons. The bilaterally different areas were found on multiple brain regions. Conclusion: The paramagnetic (positive) phase mask led to reduce signals from high susceptibility areas. To minimize partial volume effects and contributions of large vessels, the voxel-based analysis on SWI with masked non-brain components should be utilized.