• The Korean Journal of Applied Statistics
• /
• v.32 no.5
• /
• pp.693-702
• /
• 2019

Deep learning has gained popularity for the classification and prediction task. Neural network layers become deeper as more data becomes available. Saturation is the phenomenon that the gradient of an activation function gets closer to 0 and can happen when the value of weight is too big. Increased importance has been placed on the issue of saturation which limits the ability of weight to learn. To resolve this problem, Glorot and Bengio (Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, 249-256, 2010) claimed that efficient neural network training is possible when data flows variously between layers. They argued that variance over the output of each layer and variance over input of each layer are equal. They proposed a method of initialization that the variance of the output of each layer and the variance of the input should be the same. In this paper, we propose a new method of establishing initialization by adopting truncated normal distribution and truncated cauchy distribution. We decide where to truncate the distribution while adapting the initialization method by Glorot and Bengio (2010). Variances are made over output and input equal that are then accomplished by setting variances equal to the variance of truncated distribution. It manipulates the distribution so that the initial values of weights would not grow so large and with values that simultaneously get close to zero. To compare the performance of our proposed method with existing methods, we conducted experiments on MNIST and CIFAR-10 data using DNN and CNN. Our proposed method outperformed existing methods in terms of accuracy.

• Journal of the Korean Wood Science and Technology
• /
• v.40 no.6
• /
• pp.370-377
• /
• 2012

This study was conducted to investigate the color change of heat-treated wood made of Korean pine (Pinus koraiensis S. et Z.), and lodgepole pine (Pinus contorta Dougl). The wood samples were heat-treated at 9 different treatment combinations of temperature (170, 180, 200, 220, and $230^{\circ}C$), and time (90, 120, 180, 240, 270 min.) set up by a response surface method. Wood color was measured using a colorimeter, and the samples were installed for evaluating the resistance of termite in the field test sites set up in Jinju. The lightness decreased with the increase of treatment temperature and treatment time. The color differences were evaluated by the method of National Bureau of Standards (NBS), and response surface models to predict the color change were fitted as a function of treatment temperature and treatment time. The $R^2$ values of the fitted models of heat-treated Korean pine and lodgepole pines were 0.92, and 0.88, respectively, showing the potential way for the prediction of color changes. Severe damage was found from the samples in the field test site, which means that the manufactured heat-treated wood did not show the termite resistance enough for the outdoor use.

• Geotechnical Engineering
• /
• v.11 no.1
• /
• pp.9-22
• /
• 1995

The purpose of this study is to propose safety factors of pile bearing capacity based on the reliability analysis. Each prediction method involves various degrees of uncertainties. To account for these uncertainties in a systematic way, the ratios of the measured bearing capacity from pile load tests to the predicted bearing capacity are represented in the form of a probability density function. The safety factor for each design method is obtained so that the probability of pile foundation failure is less than 10-3. The Bayesian theorem is applied in a way that the distribution using static formulae is assumed to be the A-prior and the distribution using dynamic formulae or wave equation based methods is assumed to be the likelihood, and these two are combined to obtain the posterior which has the reduced uncertainty. The results of this study show that static formulae of the pile bearing capacity using the 5.p.7. N-value as well as dynamic formulae are highly unreliable and have to have the safety factor more than 7.4 : the wave equation analysis using PDA(Pile Driving Analyzer) system the most reliable with the safety factor close to 2.7. The safety factor could be reduced certain amount by adoption the Bayes methodology in pile design.

• Journal of Radiation Protection and Research
• /
• v.28 no.4
• /
• pp.311-319
• /
• 2003

We developed a grid-based Gaussian plume model to evaluate tracer release data measured at Young Gwang nuclear site in 1996. Downwind distance was divided into every 10m from 0.1km to 20km, and crosswind distance was divided into every 10m centering released point from -5km to 5km. We determined dispersion factors, ${\sigma}_y\;and\;{\sigma}_z$ using Pasquill-Gifford method computed by atmospheric stability. Forecasting ability of the grid-based Gaussian plume model was better at the 3km away from the source than 8km. We confirmed that dispersion band must be modified if receptor is far away from the source, otherwise P-G method is not appropriate to compute diffusion distance and diffusion strength in case of growing distance. So, we developed an empirical equation using linear programming. An objective function was designed to minimize sum of the absolute value between observed and computed values. As a result of application of the modified dispersion equation, prediction ability was improved rather than P-G method.

• Journal of Korean Society of Forest Science
• /
• v.106 no.2
• /
• pp.267-273
• /
• 2017

This study was conducted to construct a empirical yield table for Pinus densiflora in real forest. Since existing normal yield tables have been derived by studying and analyzing communities in ideal environment for tree growth, those tables provide more over-estimated values than ones from real forest. Because of this, there are some difficulties to apply the tables to empirical forest except for normal forest. In this study, therefore, we estimated stand growth for real forest on P. densiflora as the representative species of conifers. We used 1,957 sample plot data of P. densiflora in central Korea from National Forest Inventory (NFI) system, and analyzed through estimation, recovery and prediction in order by using Weibull function as a diameter distribution model. Weilbull and Schumacher models were applied for estimating mean DBH and mean basel area and it was found that the site index for P. densiflora in central Korea ranges from 8 to 14 at reference age 30. According to site 12 in the stand yield table, the Mean Annual Increment (MAI) of P. densiflora was $4.42m^3/ha$ at 30 years of age. Compared to existing volume table constructed before, it is showed that MAI of this study were lower. According to the paired t-test that is conducted with the gap of volume values between normal forest and real forest by site index and age, the P-value was less than 0.001 which is recognized to have a statistically significant difference. Based on the results in this study, it is considered to be helpful for practical management and management policy on P. densiflora in central Korea.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.31 no.5B
• /
• pp.449-457
• /
• 2011

Recently, increasing heavy rainfalls due to climate change and/or variability result in hydro-climatic disasters being accelerated. To cope with the extreme rainfall events in the future, hydrologic frequency analysis is usually used to estimate design rainfalls in a design target year. The rainfall data series applied to the hydrologic frequency analysis is assumed to be stationary. However, recent observations indicate that the data series might not preserve the statistical properties of rainfall in the future. This study incorporated the residual analysis and the hydrologic frequency analysis to estimate design rainfalls in a design target year considering the non-stationarity of rainfall. The residual time series were generated using a linear regression line constructed from the observations. After finding the proper probability density function for the residuals, considering the increasing or decreasing trend, rainfalls quantiles were estimated corresponding to specific design return periods in a design target year. The results from applying the method to 14 gauging stations indicate that the proposed method provides appropriate design rainfalls and reduces the prediction errors compared with the conventional rainfall frequency analysis which assumes that the rainfall data are stationary.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.36 no.8
• /
• pp.881-887
• /
• 2012

CFRP composite materials have been widely used in various fields of engineering because of their excellent properties. They show high specific stiffness and specific strength compared with metallic materiasl. Woven CFRP composite materials are fabricated from carbon fibers with two orientation angles ($0^{\circ}/90^{\circ}$), which influences the mechanical properties. Therefore, woven CFRP composite materials show different types of fracture behavior according to the load direction. Therefore, the fracture behavior of these materials needs to be evaluated according to the load direction when designing structures using these materials. In this study, we evaluate the fracture strength of plain-woven CFRP composite materials according to the load direction. We performed tests for six different angles (load direction: $0^{\circ}/90^{\circ}$, $30^{\circ}/-60^{\circ}$, $+45^{\circ}/-45^{\circ}$) and estimated the fracture strength for an arbitrary fiber angle by using the modified Tan's theory and harmonic function.

• The Journal of the Korea Contents Association
• /
• v.11 no.10
• /
• pp.388-395
• /
• 2011

This study was performed to grasp the relation between the effect and the satisfaction of a mouth gymnastics program on the promotion of oral function of old persons. The experimental group of 47 old persons conducted a mouth gymnastics exercise two times every week for 12 weeks, while the control group of 39 old persons carrying out physical examinations. The mean salivary secretion in the experimental group was 0.075 ml and more than 0.046 ml in the control group. Maximum mouth opening in the experimental group, 4.12 cm, was higher than 3.92 cm in the control. Oral diadochokinesis in experimental and control group was 1.69 times per second and 1.65, respectively. The result of principal component analysis by Quartimax method with Kaiser normalization showed that program prevalence prediction of mouth-gymnastics exercise was high, and it is estimated that this exercise program would be more popular for health promotion of old persons. From the correlation analysis, mouth gymnastics exercise made salivary flow increase(r=.592), this exercise should be recommended actively for improving mouth health of old persons. Also more modification is required for old persons with difficulty in mouth gymnastics exercise. Some movements of the mouth gymnastics exercise need be modified, and then the exercise should be publicized as one of the ways to improve holistic health of the aged.

• Journal of Advanced Marine Engineering and Technology
• /
• v.38 no.6
• /
• pp.658-665
• /
• 2014

Natural gas for marine diesel engine is considered as an important and clean source of energy because of simultaneously reducing the emission of NOx, SOx and GHG. Especially with a appearance of shale gas, the using of natural gas has been investigated aggressively and expected to expand rapidly. By the reports, gas engine and diesel engine were both in a similar performance in the power aspect, and the SFOC of gas engine was shown a little better than that of diesel engine. But the characteristics of exhaust gas emission were different according to various combustion technologies. And with lean burn technology, the emission of NOx could be reduced to 85% lower than that of diesel engine. In this paper, it was described that a simulation program has been developed to predict NOx emission. The developed program is adopted two-zone model and Wiebe function for combustion in cylinder. The effects of premixed and diffusive combustion could be simulated by using the excess air ratio as input data. And it was confirmed that the results of simulation were agreed with the general trends of exhaust gas emission according to various combustion conditions such as lean burn, premixed and diffusive combustion.

• Journal of KIISE:Databases
• /
• v.34 no.2
• /
• pp.133-140
• /
• 2007

Flash memory has penetrated our life in various forms. For example, flash memory is important storage component of ubiquitous computing or mobile products such as cell phone, MP3 player, PDA, and portable storage kits. Behind of the wide acceptance as memory is many advantages of flash memory: for instances, low power consumption, nonvolatile, stability and portability. In addition to mentioned strengths, the recent development of gigabyte range capacity flash memory makes a careful prediction that the flash memory might replace some of storage area dominated by hard disks. In order to have overwriting function, one block must be erased before overwriting is performed. This difference results in the cost of reading, writing and erasing in flash memory[1][5][6]. Since this difference has not been considered in traditional buffer replacement technologies adopted in system software such as OS and DBMS, a new buffer replacement strategy becomes necessary. In this paper, a new buffer replacement strategy, reflecting difference I/O cost and applicable to flash memory, suggest and compares with other buffer replacement strategies using workloads as Zipfian distribution and real data.