• Journal of Intelligence and Information Systems
• /
• v.23 no.2
• /
• pp.1-17
• /
• 2017

The deep learning framework is software designed to help develop deep learning models. Some of its important functions include "automatic differentiation" and "utilization of GPU". The list of popular deep learning framework includes Caffe (BVLC) and Theano (University of Montreal). And recently, Microsoft's deep learning framework, Microsoft Cognitive Toolkit, was released as open-source license, following Google's Tensorflow a year earlier. The early deep learning frameworks have been developed mainly for research at universities. Beginning with the inception of Tensorflow, however, it seems that companies such as Microsoft and Facebook have started to join the competition of framework development. Given the trend, Google and other companies are expected to continue investing in the deep learning framework to bring forward the initiative in the artificial intelligence business. From this point of view, we think it is a good time to compare some of deep learning frameworks. So we compare three deep learning frameworks which can be used as a Python library. Those are Google's Tensorflow, Microsoft's CNTK, and Theano which is sort of a predecessor of the preceding two. The most common and important function of deep learning frameworks is the ability to perform automatic differentiation. Basically all the mathematical expressions of deep learning models can be represented as computational graphs, which consist of nodes and edges. Partial derivatives on each edge of a computational graph can then be obtained. With the partial derivatives, we can let software compute differentiation of any node with respect to any variable by utilizing chain rule of Calculus. First of all, the convenience of coding is in the order of CNTK, Tensorflow, and Theano. The criterion is simply based on the lengths of the codes and the learning curve and the ease of coding are not the main concern. According to the criteria, Theano was the most difficult to implement with, and CNTK and Tensorflow were somewhat easier. With Tensorflow, we need to define weight variables and biases explicitly. The reason that CNTK and Tensorflow are easier to implement with is that those frameworks provide us with more abstraction than Theano. We, however, need to mention that low-level coding is not always bad. It gives us flexibility of coding. With the low-level coding such as in Theano, we can implement and test any new deep learning models or any new search methods that we can think of. The assessment of the execution speed of each framework is that there is not meaningful difference. According to the experiment, execution speeds of Theano and Tensorflow are very similar, although the experiment was limited to a CNN model. In the case of CNTK, the experimental environment was not maintained as the same. The code written in CNTK has to be run in PC environment without GPU where codes execute as much as 50 times slower than with GPU. But we concluded that the difference of execution speed was within the range of variation caused by the different hardware setup. In this study, we compared three types of deep learning framework: Theano, Tensorflow, and CNTK. According to Wikipedia, there are 12 available deep learning frameworks. And 15 different attributes differentiate each framework. Some of the important attributes would include interface language (Python, C ++, Java, etc.) and the availability of libraries on various deep learning models such as CNN, RNN, DBN, and etc. And if a user implements a large scale deep learning model, it will also be important to support multiple GPU or multiple servers. Also, if you are learning the deep learning model, it would also be important if there are enough examples and references.

• Journal of Intelligence and Information Systems
• /
• v.17 no.4
• /
• pp.241-254
• /
• 2011

Financial time-series forecasting is one of the most important issues because it is essential for the risk management of financial institutions. Therefore, researchers have tried to forecast financial time-series using various data mining techniques such as regression, artificial neural networks, decision trees, k-nearest neighbor etc. Recently, support vector machines (SVMs) are popularly applied to this research area because they have advantages that they don't require huge training data and have low possibility of overfitting. However, a user must determine several design factors by heuristics in order to use SVM. For example, the selection of appropriate kernel function and its parameters and proper feature subset selection are major design factors of SVM. Other than these factors, the proper selection of instance subset may also improve the forecasting performance of SVM by eliminating irrelevant and distorting training instances. Nonetheless, there have been few studies that have applied instance selection to SVM, especially in the domain of stock market prediction. Instance selection tries to choose proper instance subsets from original training data. It may be considered as a method of knowledge refinement and it maintains the instance-base. This study proposes the novel instance selection algorithm for SVMs. The proposed technique in this study uses genetic algorithm (GA) to optimize instance selection process with parameter optimization simultaneously. We call the model as ISVM (SVM with Instance selection) in this study. Experiments on stock market data are implemented using ISVM. In this study, the GA searches for optimal or near-optimal values of kernel parameters and relevant instances for SVMs. This study needs two sets of parameters in chromosomes in GA setting : The codes for kernel parameters and for instance selection. For the controlling parameters of the GA search, the population size is set at 50 organisms and the value of the crossover rate is set at 0.7 while the mutation rate is 0.1. As the stopping condition, 50 generations are permitted. The application data used in this study consists of technical indicators and the direction of change in the daily Korea stock price index (KOSPI). The total number of samples is 2218 trading days. We separate the whole data into three subsets as training, test, hold-out data set. The number of data in each subset is 1056, 581, 581 respectively. This study compares ISVM to several comparative models including logistic regression (logit), backpropagation neural networks (ANN), nearest neighbor (1-NN), conventional SVM (SVM) and SVM with the optimized parameters (PSVM). In especial, PSVM uses optimized kernel parameters by the genetic algorithm. The experimental results show that ISVM outperforms 1-NN by 15.32%, ANN by 6.89%, Logit and SVM by 5.34%, and PSVM by 4.82% for the holdout data. For ISVM, only 556 data from 1056 original training data are used to produce the result. In addition, the two-sample test for proportions is used to examine whether ISVM significantly outperforms other comparative models. The results indicate that ISVM outperforms ANN and 1-NN at the 1% statistical significance level. In addition, ISVM performs better than Logit, SVM and PSVM at the 5% statistical significance level.

• Journal of dental hygiene science
• /
• v.16 no.3
• /
• pp.217-224
• /
• 2016

The aim of this study was to evaluate whether the number of existing permanent teeth is associated with chronic obstructive pulmonary disease (COPD) in a representative sample of Korean adults. Data from 3,107 subjects who participated in the 2009 Korea National Health and Nutrition Examination Survey were examined. The dependent variable was COPD and the independent variable was the number of existing permanent teeth. Spirometry results were classified into three groups (normal pattern, restrictive pattern, and obstructive pattern) by trained technicians. We used dichotomized COPD variables (no vs. yes). The number of existing permanent teeth was evaluated by oral examination and divided into 3 groups (0~19, 20~27, and 28). Demographic factors (age group and sex group), socioeconomic status (education and income), health behaviors (smoking and drinking), oral health and behavior (frequency of toothbrushing; periodontitis; decayed, missing, filled, permanent teeth index; and denture status), and general health status (body mass index, diabetes mellitus, and hypertension) were included as confounders in the analysis. Bivariate analysis and multivariate logistic regression analyses including confounders were applied, and all analyses considered a complex sampling design. Stratified analysis was performed by smoking status. After controlling for various confounders, there was a significant association between the number of existing permanent teeth and COPD (odds ratio [OR], 1.90; 95% confidence interval [CI], 1.20~3.00 for the 20~27 group; OR, 3.93; 95% CI, 1.75~8.84 for the 0~19 group). The association was more significant in current smokers (OR, 8.90; 95% CI, 2.53~31.33). Our data indicate that the number of existing permanent teeth was independently associated with COPD, especially in current smokers. Further longitudinal research is needed to determine whether oral health promotion plays a role in the improvement of lung function and prevention of COPD.

• Tuberculosis and Respiratory Diseases
• /
• v.80 no.3
• /
• pp.277-283
• /
• 2017

Background: Smoking cessation is the most powerful intervention to modify progress of chronic obstructive pulmonary disease (COPD), and nicotine dependence is one of the most important determinants of success or failure in smoking cessation. We evaluated nicotine dependence status and investigated factors associated with moderate to high nicotine dependence in patients with COPD. Methods: We included 53 current smokers with COPD in the Korean Obstructive Lung Disease II cohort enrolled between January 2014 and March 2016. Nicotine dependence was measured by using Fagerstrom test for nicotine dependence (FTND). Cognitive function was assessed by Korean version of Montreal Cognitive Assessment. Results: The median FTND score was 3, and 32 patients (60%) had moderate to high nicotine dependence. The median smoking amount was 44 pack-years, which was not related to nicotine dependence. Multiple logistic regression analysis revealed that high education status (odds ratio, 1.286; 95% confidence interval, 1.036-1.596; p=0.023), age <70 (odds ratio, 6.407; 95% confidence interval, 1.376-29.830; p=0.018), and mild to moderate airflow obstruction (odds ratio, 6.969; 95% confidence interval, 1.388-34.998; p=0.018) were related to moderate to high nicotine dependence. Conclusion: Nicotine dependence does not correlate with smoking amount, but with education level, age, and severity of airflow obstruction. Physicians should provide different strategies of smoking cessation intervention for current smokers with COPD according to their education levels, age, and severity of airflow obstruction.

• Journal of Genetic Medicine
• /
• v.5 no.2
• /
• pp.105-110
• /
• 2008

Purpose: Preecalmpsia is a pregnancy-specific disorder that reflects widespread endothelial dysfunction resulting from increases of adhesion molecule expression. Intercellular adhesion molecule-1 (ICAM-1) is involved in the pathogenetic mechanisms responsible for preeclampsia, and ICAM-1 plasma levels and/or function is genetically influenced. Therefore, we evaluated the distribution of ICAM-1 gene K469E polymorphism in pregnant Korean women with preeclampsia and evaluated the association between this polymorphism and preeclampsia. Methods: The K469E polymorphism was analyzed in peripheral blood samples from 197 preeclamptic pregnancies and 193 normotensive pregnancies by a SNapShot kit and an ABI Prism 3100 Genetic analyzer. Results: Genotypic and allelic frequencies of ICAM-1 gene polymorphism (K469E) did not differ between preeclamptic and normotensive pregnancies. The distributions of the KK, KE, and EE genotypes were 40.6%, 43.7%, and 15.7%, respectively, in preeclamptic pregnancies and 38.9%, 45.1%, and 16.1%, respectively, in normotensive pregnancies. The frequencies of K and E alleles were 0.62 and 0.38, respectively, in preeclamptic pregnancies and 0.61 and 0.39, respectively, in normotensive pregnancies. By multiple logistic regression analysis, there was no increased risk of preeclampsia in subjects with ICAM-1 KE (OR, 1.08; P=0.74) or EE (OR, 1.07; P=0.88) genotypes. Conclusion: This study suggests that the ICAM-1 gene K469E polymorphism does not associate with an increased risk of preeclampsia in pregnant Korean women.

• Journal of the Korean Society of Fisheries and Ocean Technology
• /
• v.40 no.1
• /
• pp.86-94
• /
• 2004

This study was conducted to mesh selectivity of Beam trawl for shrimps fishing experiment in the coastal waters around Geomundo, South sea of Korea, during from Oct. to Nov. 2002. The selectivity parameters of big head shrimp (Solenocera melantho) have been studied on the covered con-end method. with mesh of 8, 38, 51 and 61 mm. Selection curves and selection parameters were calculated by using a logistic function S=1/(1+exp-(aCL+b)). The mesh selection master curves were estimated by S=1/(1+exp$^{({\alpha}(CL/M)+{\beta}}$), and the optimum mesh size were calculated with (L/M)50 of master curve. Optimum mesh size and selectivity master curves for the southern rough shrimp (Yrachysalambria curvirostris) and smoothshell shrimp (Parapenaeopsis tenella) optimum mesh size and selectivity master curves were estimated by big head shrimp master curves. The results obtained are summarized as follows : Selection parameters '${\alpha}$' and '${\beta}$' of the master curve for big head shrimp were 8.84 and -5.89, and The selection factor of the master curve (L/M)$_{50}$ was 0.67. The optimum mesh size of minimum length for sexual maturity for big head shrimp was 30.7 mm. Estimated (L/M)$_{50}$ for southern rough shrimp and smoothshell shrimp by using the master curve of big head shrimp was 0.73 and the optimum mesh sizes were 25.5 mm for southern rough shrimp and 16.9 mm for smoothshell shrimp, respectively.

• The Journal of Korean Physical Therapy
• /
• v.14 no.3
• /
• pp.238-272
• /
• 2002

This study was intended to find out the factors affecting to the IADL, allowing objective assessment of physical function status of increasing elderly populations. The subjects of 635 elderly persons aged over 65 years old who live in Taejon metropolitan city were interviewed during the two-month period from June to July of 2000. The IADL of Older American Resources and Services(OARS), developed in Duke medical college of USA, was interpreted for use. The study results were statistically processed using SPSSWIN(ver 10.0) and conferred the following results: 1. Among the seven items of IADL, the women showed higher rates of 'yes' in items about using the telephone, getting to the places out of walking distance, going shopping for groceries, taking their own medicine while outnumbered by the men only in the item about managing their own money, without significant differences between them in items about preparing their own meals and doing their own housework. 2. In terms of IADL scores, $82.0\%$ of subjects showed the normal range of scores and $18.0\%$ were under the normal range, meaning dysfunction IADL. Specifications of general characteristics revealed that more dysfunctional were the male subjects, the higher age groups who are more than 75 years old, the higher educated groups, the groups who live in nursing homes, the groups without a spouse. 3. Based on daily activities, lower scores of IADL were found in the subjects who don't go out, who don't have recreational activities, who don't attend elderly gatherings in their neighborhood, who don't hold social meetings. Specifications of psychological traits revealed that more dysfunctional were IADL in the subjects who don't feel satisfied with their lives of the past or the present, who have a deep feeling of isolation, and who don't have the will to live. 4. Among 7 items of health habits, only the subjects who don't regular exercise had lower scores of IADL than those who do. According to HPI, the lower HPI, the more dysfunctional. 5. Based on the factors associated with IADL, the odds ratio of the subjects who don't live together with their families were 1.53 times that of the ones who do, who are educated 3.22 times that of the ones who are not, who don't have spouses 2.09 times that of the ones who do, who don't go out 4.35 times that of the ones who frequently go out, who don't recreational activities for an interest 2.64 times that of the ones who do, who don't attend elderly gatherings in their neighborhood 1.47 times that of the ones who do, who don't hold social meetings 2.23 times that of the ones who do, who don't feel satisfied with their present living 1.43 times that of the ones who do, who have a feeling of isolation 1.53times that of the ones who don'1, who have the weak will to live 3.21 times that of the ones who have the strong one, and who don't regular exercise 2.45 times that of the ones who do. 6. Logistic regression analysis of the study results found that such factors are significantly related as the degree of education, with/without spouse, social meetings, the will to live, and regular exercise, and that higher rates of dysfunctional subjects were in the more educated group, in the group without spouse, in the group who don't frequently go out, who don't have social meetings, who have the weak will to live, and who don't exercise.

• Journal of the Korean Society of Fisheries and Ocean Technology
• /
• v.29 no.4
• /
• pp.247-259
• /
• 1993

The fishing experiment was carried out by the training ship Saebada in order to analyse the mesh selectivity for trawl cod-end, in the Southern Korea Sea and the East China Sea from June. 1991 through August, 1992. The trawl cod-end used in this experiment has the trouser type of cod-end with cover net. and the mesh selectivity was examined for the five kinds of the opening of mesh in its cod-end part. A total of 163 hauls, of which having mesh size 51.2mm ; A 89, 70.2mm ; B 54, 77.6mm ; C 55, 88.0mm ; D 52 and 111.3mm ; E 20 were used respectively. Selection curves and selection parameters were calculated by using a logistic function, S=1/(1+exp super(-(aL+b)) ). The mesh election master curves were estimated by S=1/(1+exp super(-[a(L/M)+$\beta$]) ). and the optimum mesh size were calculated with (L/M) sub(50) of master curve. In these cases 'a' and '$\alpha$' are slope, 'b' and '$\beta$' are intercept. 'L' is body length of the target species of fishes, 'M' is the mesh size, and 'S' denotes mesh selectivity. In this report, the four species of compressed form fishes were taken analized according to fish shape, and 'S' denotes mesh selectivity. In this report, the four species of compressed form fishes were taken analized according to fish shape, and the results obtained are summarized as follows: 1. Red seabream Pagrus major(Temminct et Schlegel) and yellow porgy Dentex tumifrons(Temminct et Schlegel) ; Selection rate in each mesh size of A, B, C, D and E were 99.7%, 97.5%, 91.4%, 76.7% and 57.8% respectively. Selection parameters 'a' and 'b' of mesh sizes C, D and E were 2.65 and -28.62, 4.40 and -77.73, 2.31 and -46.99, and their selection factors were 1.39, 2.10, 1.83 respectively. Selection parameters of master curve '$\alpha$' and '$\beta$' were 3.05 and -5.65 respectively, and (L/M) sub(50) was 1.85. The optimum mesh size of Red seabream was 141mm. 2. Filefish Thamnaconus modestus (Gunther) ; Selection rate in each mesh size of A, B, C, D and E were 99.6%, 98.3%, 91.2%, 80.0% and 48.6% respectively. Selection parameters 'a' and 'b' of mesh sizes C, D and E were 5.82 and -55.10, 2.92 and -36.90, 3.91 and -63.09, and their selection factors were 1.35, 1.44, 1.45 respectively. Selection parameters of master curve '$\alpha$' and '$\beta$' were 3.02 and -4.32 respectively, and (L/M) sub(50) was 1.43. The optimum mesh size was 129mm. 3. Target dory Zeus faber Valenciennes ; Selection rate in each mesh size of A, B, C, D and E were 99.7%, 100%, 83.2%, 91.6% and 65.0% respectively. Selection parameters 'a' and 'b' of mesh sizes C, D and E were 3.85 and -32.46, 4.19 and -57.38, 2.45 and -40.03, and their selection factors were 1.09, 1.56, 1.47 respectively. Selection parameters of master curve '$\alpha$' and '$\beta$' were 2.64 and -3.53 respectively, and (L/M) sub(50) was 1.34. The optimum mesh size was 127mm. 4. Butterfish Psenopsis anomala (Temminct et Schlegel) ; Selection rate in each mesh size of A, B, C, D and E were 99.2%, 34.1%, 46.5%, 14.3% and 2.4% respectively. Selection parameters 'a' and 'b' of mesh sizes B, C and D were 5.35 and -71.70, 5.07 and -69.25, 3.31 and -62.06 and their selection factors were 1.91, 1.75, 2.13 respectively. Selection parameters of master curve '$\alpha$' and '$\beta$' were 3.16 and -6.24 respectively, and (L/M) sub(50) was 1.98. The optimum mesh size was 71mm.

• Tuberculosis and Respiratory Diseases
• /
• v.62 no.1
• /
• pp.11-18
• /
• 2007

Background: Patients with chronic obstructive pulmonary disease (COPD) have often been reported to suffer from depression and anxiety possibly due to the exacerbation, hospitalization and mortality of COPD. However,scarce data are available in Korea. This study assessed degree of depression and anxiety, and evaluated the factors associated with depressive symptoms in COPD. Methods: The cross-sectional data on the lung function measurements, smoking behavior, body mass index (BMI), age, gender, depressive symptoms using Beck Depression Inventory (BDI) and anxiety using the State-Trait Anxiety Inventory (STAI) were evaluated in 72 outpatients with COPD and 50 controls without underling lung diseases from September, 2005 to October, 2006 in the Ewha medical center. Results: 1) The age, body mass index (BMI) and serum albumin levels were similar in the patients and controls. The BDI scores (16(0-37) vs. 12(1-30), p=0.001) and the prevalence of depression (36% vs. 6%, p<0.0001) were higher in the COPD patients than in the controls. In the COPD group, the prevalence of depression increased with increasing GOLD stage (p=0.008). The prevalence was 18%(4/22), in mild cases, 30%(6/20) in moderate cases, 52%(13/25) in severe cases and 60%(3/5) in very severe cases. 2) The SAI and TAI scores were higher in the COPD patients (44(20-67) and 47(20-66)) than in the healthy controls (39(26-65) and 44(33-90)). There were a significant correlation between the depression and anxiety scores (p<0.001). 3) A lower BMI, lower postbronchodilator $FEV_1$, current smoking behavior and severity of COPD were univariately associated with the depressive group in COPD, 4) while multivariate logistic analysis revealed only the severe-to-very severe group (OR 3.9, 95% CI 1.2 to 12.9) to be independently associated with depressive symptoms. Conclusion: COPD is strongly associated with depression and anxiety. Therfore, screening for psychological problems in COPD patients is essential, particularly in patients with severe-to-very severe COPD.

• Tuberculosis and Respiratory Diseases
• /
• v.45 no.4
• /
• pp.746-753
• /
• 1998

Background : Cigarette smoking is closely related to both lung cancer and chronic obstructive pulmonary disease. The incidence of lung cancer is higher in patients with obstructive ventilatory impairment than in patients without obstructive ventilatory impairment regardless of smoking. So, obstructive ventilatory impairment is suspected as an independent risk factor of lung cancer. Methods: For the evaluation of the role of obstructive ventilatory impairment as a risk factor of lung cancer, a total of 73 cases comprising 47 cases of malignant and 26 benign solitary pulmonary nodule were analyzed retrospectively. A comparative study of analysis of forced expiratory volume curves and frequencies of obstructive ventilatory impairment were made between cases with malignant and benign nodules. Results: In comparison of vital capacity and parameters derived from forced expiratory volume curve between two groups. VC, FVC and $FEV_1$ were not significantly different. whereas $FEV_1/FVC%$ and FEF 25-75% showed a significant decrease in the cases with malignant nodule. The frequency of obstructive ventilatory impairment determined by pulmonary function test was significantly higher in the cases with malignant nodule(23.4%) than in benign nodule(3.8%). When the risk for lung cancer was examined by the presence or absence of obstructive ventilatory impairment using the logistic regression analysis, the unadjusted relative risk for the lung cancer of obstructive ventilatory impairment was 17.17. When the effect of smoking and age were considered, the relative risk was to 8.13. Conclusion: These findings suggest that an obstructive ventilatory impairment is a risk factor of lung cancer.