• Journal of Intelligence and Information Systems
• /
• v.23 no.2
• /
• pp.71-88
• /
• 2017

Language models were originally developed for speech recognition and language processing. Using a set of example sentences, a language model predicts the next word or character based on sequential input data. N-gram models have been widely used but this model cannot model the correlation between the input units efficiently since it is a probabilistic model which are based on the frequency of each unit in the training set. Recently, as the deep learning algorithm has been developed, a recurrent neural network (RNN) model and a long short-term memory (LSTM) model have been widely used for the neural language model (Ahn, 2016; Kim et al., 2016; Lee et al., 2016). These models can reflect dependency between the objects that are entered sequentially into the model (Gers and Schmidhuber, 2001; Mikolov et al., 2010; Sundermeyer et al., 2012). In order to learning the neural language model, texts need to be decomposed into words or morphemes. Since, however, a training set of sentences includes a huge number of words or morphemes in general, the size of dictionary is very large and so it increases model complexity. In addition, word-level or morpheme-level models are able to generate vocabularies only which are contained in the training set. Furthermore, with highly morphological languages such as Turkish, Hungarian, Russian, Finnish or Korean, morpheme analyzers have more chance to cause errors in decomposition process (Lankinen et al., 2016). Therefore, this paper proposes a phoneme-level language model for Korean language based on LSTM models. A phoneme such as a vowel or a consonant is the smallest unit that comprises Korean texts. We construct the language model using three or four LSTM layers. Each model was trained using Stochastic Gradient Algorithm and more advanced optimization algorithms such as Adagrad, RMSprop, Adadelta, Adam, Adamax, and Nadam. Simulation study was done with Old Testament texts using a deep learning package Keras based the Theano. After pre-processing the texts, the dataset included 74 of unique characters including vowels, consonants, and punctuation marks. Then we constructed an input vector with 20 consecutive characters and an output with a following 21st character. Finally, total 1,023,411 sets of input-output vectors were included in the dataset and we divided them into training, validation, testsets with proportion 70:15:15. All the simulation were conducted on a system equipped with an Intel Xeon CPU (16 cores) and a NVIDIA GeForce GTX 1080 GPU. We compared the loss function evaluated for the validation set, the perplexity evaluated for the test set, and the time to be taken for training each model. As a result, all the optimization algorithms but the stochastic gradient algorithm showed similar validation loss and perplexity, which are clearly superior to those of the stochastic gradient algorithm. The stochastic gradient algorithm took the longest time to be trained for both 3- and 4-LSTM models. On average, the 4-LSTM layer model took 69% longer training time than the 3-LSTM layer model. However, the validation loss and perplexity were not improved significantly or became even worse for specific conditions. On the other hand, when comparing the automatically generated sentences, the 4-LSTM layer model tended to generate the sentences which are closer to the natural language than the 3-LSTM model. Although there were slight differences in the completeness of the generated sentences between the models, the sentence generation performance was quite satisfactory in any simulation conditions: they generated only legitimate Korean letters and the use of postposition and the conjugation of verbs were almost perfect in the sense of grammar. The results of this study are expected to be widely used for the processing of Korean language in the field of language processing and speech recognition, which are the basis of artificial intelligence systems.

• Journal of Intelligence and Information Systems
• /
• v.18 no.2
• /
• pp.29-45
• /
• 2012

Bond rating is regarded as an important event for measuring financial risk of companies and for determining the investment returns of investors. As a result, it has been a popular research topic for researchers to predict companies' credit ratings by applying statistical and machine learning techniques. The statistical techniques, including multiple regression, multiple discriminant analysis (MDA), logistic models (LOGIT), and probit analysis, have been traditionally used in bond rating. However, one major drawback is that it should be based on strict assumptions. Such strict assumptions include linearity, normality, independence among predictor variables and pre-existing functional forms relating the criterion variablesand the predictor variables. Those strict assumptions of traditional statistics have limited their application to the real world. Machine learning techniques also used in bond rating prediction models include decision trees (DT), neural networks (NN), and Support Vector Machine (SVM). Especially, SVM is recognized as a new and promising classification and regression analysis method. SVM learns a separating hyperplane that can maximize the margin between two categories. SVM is simple enough to be analyzed mathematical, and leads to high performance in practical applications. SVM implements the structuralrisk minimization principle and searches to minimize an upper bound of the generalization error. In addition, the solution of SVM may be a global optimum and thus, overfitting is unlikely to occur with SVM. In addition, SVM does not require too many data sample for training since it builds prediction models by only using some representative sample near the boundaries called support vectors. A number of experimental researches have indicated that SVM has been successfully applied in a variety of pattern recognition fields. However, there are three major drawbacks that can be potential causes for degrading SVM's performance. First, SVM is originally proposed for solving binary-class classification problems. Methods for combining SVMs for multi-class classification such as One-Against-One, One-Against-All have been proposed, but they do not improve the performance in multi-class classification problem as much as SVM for binary-class classification. Second, approximation algorithms (e.g. decomposition methods, sequential minimal optimization algorithm) could be used for effective multi-class computation to reduce computation time, but it could deteriorate classification performance. Third, the difficulty in multi-class prediction problems is in data imbalance problem that can occur when the number of instances in one class greatly outnumbers the number of instances in the other class. Such data sets often cause a default classifier to be built due to skewed boundary and thus the reduction in the classification accuracy of such a classifier. SVM ensemble learning is one of machine learning methods to cope with the above drawbacks. Ensemble learning is a method for improving the performance of classification and prediction algorithms. AdaBoost is one of the widely used ensemble learning techniques. It constructs a composite classifier by sequentially training classifiers while increasing weight on the misclassified observations through iterations. The observations that are incorrectly predicted by previous classifiers are chosen more often than examples that are correctly predicted. Thus Boosting attempts to produce new classifiers that are better able to predict examples for which the current ensemble's performance is poor. In this way, it can reinforce the training of the misclassified observations of the minority class. This paper proposes a multiclass Geometric Mean-based Boosting (MGM-Boost) to resolve multiclass prediction problem. Since MGM-Boost introduces the notion of geometric mean into AdaBoost, it can perform learning process considering the geometric mean-based accuracy and errors of multiclass. This study applies MGM-Boost to the real-world bond rating case for Korean companies to examine the feasibility of MGM-Boost. 10-fold cross validations for threetimes with different random seeds are performed in order to ensure that the comparison among three different classifiers does not happen by chance. For each of 10-fold cross validation, the entire data set is first partitioned into tenequal-sized sets, and then each set is in turn used as the test set while the classifier trains on the other nine sets. That is, cross-validated folds have been tested independently of each algorithm. Through these steps, we have obtained the results for classifiers on each of the 30 experiments. In the comparison of arithmetic mean-based prediction accuracy between individual classifiers, MGM-Boost (52.95%) shows higher prediction accuracy than both AdaBoost (51.69%) and SVM (49.47%). MGM-Boost (28.12%) also shows the higher prediction accuracy than AdaBoost (24.65%) and SVM (15.42%)in terms of geometric mean-based prediction accuracy. T-test is used to examine whether the performance of each classifiers for 30 folds is significantly different. The results indicate that performance of MGM-Boost is significantly different from AdaBoost and SVM classifiers at 1% level. These results mean that MGM-Boost can provide robust and stable solutions to multi-classproblems such as bond rating.

• 한국석유지질학회:학술대회논문집
• /
• autumn
• /
• pp.63-65
• /
• 2001

The Chosen Supergroup (Cambro-Ordovician), mid-east Korea consists mainly of shallow marine carbonates and contains a variety of limestone conglomerates. These conglomerates largely comprise oligomictic, rounded lime-mudstone clasts of various size and shape (equant, oval, discoidal, tabular, and irregular) and dolomitic shale matrices. Most clasts are characterized by jigsaw-fit (mosaic), disorganized, or edgewise fabric and autoclastic lithology. Each conglomerate layer is commonly interbedded with limestone-dolomitic shale couplets and occasionally underlain by fractured limestone layer, capped by calcareous shale. According to composition, characteristic sedimentary structures, and fabric, limestone conglomerates in the Hwajol, Tumugol, Makkol, and Mungok formations of Chosen Supergroup can be classified into 4 types: (1) disorganized polymictic conglomerate (Cd), (2) horizontally stratified polymictic conglomerate (Cs), (3) mosaic conglomerate (Cm), and (4) disorganized/edgewise oligomictic conglomerate (Cd/e). These conglomerates are either depositional (Cd and Cs) or diagenetic (Cm and Cd/e) in origin. Depositional conglomerates are interpreted as storm deposits, tidal channel fills, or transgressive lag deposits. On the other hand, diagenetic conglomerates are not deposited by normal sedimentary processes, but formed by post-depositional diagenetic processes. Diagenetic conglomerates in the Chosen Supergroup are characterized by autoclastic and oligomictic lithology of lime-mudstone clasts, jigsaw-fit (mosaic) fabric, edgewise fabric, and a gradual transition from the underlying bed (Table 1). Autoclastic and oligomictic lithologies may be indicative of subsurface brecciation (fragmentation). Consolidation of lime-mudstone clasts pre-requisite for brecciation may result from dissolution and reprecipitation of CaCO3 by degradation of organic matter during burial. Jigsaw-fit fabric has been considered as evidence for in situ fragmentation. The edgewise fabric is most likely formed by expulsion of pore fluid during compaction. The lower boundary of intraformational conglomerates of depositional origin is commonly sharp and erosional. In contrast, diagenetic conglomerate layers mostly show a gradual transition from the underlying unit, which is indicative of progressive fragmentation upward (Fig. 1). The underlying fractured limestone layer also shows evidence for in situ fragmentation such as jigsaw-fit fabric and the same lithology as the overlying conglomerate layer (Fig, 1). Evidence from the conglomerate beds in the Chosen Supergroup suggests that diagenetic conglomerates are formed by in situ subsurface fragmentation of limestone layers and rounding of the fragments. In situ subsurface fragmentation may be primarily due to compaction, dewatering (upward-moving pore fluids), and dissolution, accompanying volume reduction. This process commonly occurs under the conditions of (1) alternating layers of carbonate-rich and carbonate-poor sediments and (B) early differential cementation of carbonate-rich layers. Differential cementation commonly takes place between alternating beds of carbonate-rich and clay-rich layers, because high carbonate content promotes cementation, whereas clay inhibits cementation. After deposition of alternating beds and differential cementation, with progressive burial, upward-moving pore fluid may raise pore-pressure in the upper part of limestone layers, due to commonly overlying impermeable shale layers (or beds). The high pore-pressure may reinforce propagation of fragmentation and cause upward-expulsion of pore fluid which probably produces edgewise fabric of tabular clasts. The fluidized flow then extends laterally, causing reorientation and further rounding of clasts. This process is analogous to that of autobrecciation, which can be analogously termed autoconglomeration. This is a fragmentation and rounding process whereby earlier semiconsolidated portions of limestone are incorporated into still fluid portions. The rounding may be due mainly to immiscibility and surface tension of lime-mud. The progressive rounding of the fragmented clasts probably results from grain attrition by fluidized flow. A synthetic study of limestone conglomerate beds in the Chosen Supergroup suggests that very small percent of the conglomerate layers are of depositional origin, whereas the rest, more than $80\%$, are of diagenetic origin. The common occurrence of diagenetic conglomerates warrants further study on limestone conglomerates elsewhere in the world.

• Clinics in Shoulder and Elbow
• /
• v.16 no.1
• /
• pp.17-26
• /
• 2013

Purpose: A comparison of the radiographic and the clinical outcomes between two different surgical approaches-Deltoid splitting and Delto-pectoral interval-on the proximal humerus fractures treated by locking compressive plate (LCP), is done. Materials and Methods: Medical records and pre- and postoperative radiographs were reviewed retrospectively for 75 adult patients who underwent surgical fixations with locking compressive plates from May 2005 to December 2011. Patients were divided into two groups according to the surgical methods. Differences in the neck-shaft angle between immediate postoperative period and final follow-up were compared between the two groups. Differences in constant score and Korean shoulder score (KSS) between affected arms and contralateral arms at final follow-up were also compared. Results: The differences in the neck-shaft angle between immediate postoperative period and at final follow-up was 12.04 degrees on average in Deltoid splitting approach and 10.20 degrees in Delto-pectoral interval approach, which was not statistically significant. Differences in constant score/KSS between the affected arm and the contralateral arm were 13.78/22.74 points in deltoid-splitting approach on average and 19.41/31.13 points in Delto-pectoral interval approach, showing that deltoid-splitting approach is significantly superior. Conclusion: Deltoid-splitting approach showed better functional outcomes in the fracture reduction and internal fixation using LCP for the treatment of unstable proximal humerus fractures.

• The korean journal of orthodontics
• /
• v.27 no.3 s.62
• /
• pp.411-420
• /
• 1997

The propose of this study was to quantify the changes of soft tissue profile following orthodontic treatment and to evaluate the relationship of those to the skeletal elements. Pre-and post-treatment lateral cephalometric head films of 40 cases(20 extraction cases, 20 non-extraction cases) were traced, and the changes following treatment were measured and quantified by digital subtraction method, and statisticall analyzed. The obtained results were as follows; 1. in extraction group, the change of upper lip area(UL) was $558.60\pm355.17$ pixels, that of lower lip area(LL) was $941.15\pm364.07$ pixels. But, in non-extraction group the change of uper lip area(UL) was $125.65\pm404.16$ pixels, that of lower lip area(LL) was $104.05\pm440.93$ pixels, which was significantly lesser than those in extraction group. 2. In extraction group, there was significant correlationship between upper lip area change(UL) and difference of upper incisor point(${\Delta}UIP$). Lower lip area change(LL) was significantly correlated with difference of upper incisor(${\Delta}UIP$), difference of Franlrfort upper incisor angle(${\Delta}FUIA$) or difference of interincisal angle(${\Delta}IIA$). 3. In extraction group, the ratio of difference of upper incisor point(${\Delta}UIP$) to difference of labrale superius(${\Delta}LSP$) was 1.68; difference of lower incisor point(${\Delta}LIP$) to difference of labrale inferius(${\Delta}LI$) was 1.19; difference of upper incisor point(${\Delta}UIP$) to increment in upper lip thickness(${\Delta}TUL$) was 1.95. 4. In non-extraction group, there was a significant correlationship between upper lip area change(UL) and difference of upper incisor point(${\Delta}UIP$).

• Asian-Australasian Journal of Animal Sciences
• /
• v.33 no.3
• /
• pp.480-489
• /
• 2020

Objective: Sow milk (SM) may not be able to meet the piglet's nutritional needs in late lactation. Hence, this study was conducted to investigate the effects of early commercial milk (CM) supplement on the mucosal morphology, bacterial community and bacterial metabolites in jejunum of piglets. Methods: Ten litters of newborn piglets ([Yorkshire×Landrace]×Duroc) were randomly divided into 2 groups of 5 litters. The piglets in the control group were suckled by the sow (SM), while the piglets in the treatment group (CM supplement) were supplemented with a CM supplement along with suckling from d 4 to d 28 of age. Results: No significant differences were observed about jejunal mucosal morphology on d 28 and d 35 between two groups. On d 28, the activity of lactase in the jejunum was significantly decreased in the CM group, while the activity of sucrase and the ratio of maltase to lactase were significantly increased (p<0.05). On d 35, the activity of maltase in the jejunum was significantly increased in the CM group (p<0.05), and maltase to lactase ratio tended to increase in the CM group (p = 0.065). In addition, piglets in the CM group had a higher abundance of Clostridium XI, Tuicibater, and Moraxella in the jejunum on d 28, while the abundance of Lactobacillus was significantly increased on d 35 (p<0.05). Conclusion: The early CM supplement improved the maturation of the jejunum to some extent by enhancing the maltase and sucrase activities. Moreover, the early CM supplement could help maintain the homeostasis of internal environment in jejunum by increasing the microbial-derived metabolites.

• Journal of The Korean Association For Science Education
• /
• v.22 no.4
• /
• pp.796-805
• /
• 2002

This study examined the knowledge structure constructed by children before formal instruction, and successive changes in the structural complexity of knowledge during and after the learning of 'Structure and Function of Plant' unit. It also investigated how those changes were affected by children's learning modes. The researchers made the 5th graders draw the first draft of their concept map to see the pre-existing knowledge structure concerned with the unit and four more concept maps after completing every fourth lesson. And to see how long their knowledge structures were preserved, the researchers made children draw additional concept maps in 3 days, 3 months, and 7 months after completing the unit. Children drew their current concept maps on the basis of the previous one while learning the unit and without the previous one after completing the unit. Each concept map drawn by children showed the degree of their current understanding on the structures and functions of plants. The results revealed that only two levels of hierarchy and five relationships among the components of the first concept map(relationship, hierarchy, cross link and example) were proven to be valid in terms of conceptual relevance. Growth in the structural complexity of knowledge took place progressively throughout the unit and the effects of learning mode on the growth were favorably reflected in concept map scores of meaningful learners over time(relationship, cross link, example: p<.01, hierarchy: p<.05). Although there were some differences on the concept map scores between two types of learners, they commonly showed that knowledge restructuring had occurred apparently in the early periods from the 1st to the 6th lesson and had not occurred at all in the last period of the unit. The frequency of tuning was higher in rote learners than in meaningful learners throughout the unit, but the frequency of accretion was reverse. Concept map scores of rote learners constructed in the course of learning of the unit decreased little by little gradually in all the categories after completing the unit. However, the average total map score of meaningful learners increased a little more in 7 months than in 3 months after completing the unit. Therefore it can be inferred that meaningful learners construct more stable and well-differentiated knowledge structures than the rote learners.

• Journal of Chest Surgery
• /
• v.38 no.5 s.250
• /
• pp.357-365
• /
• 2005

Background: Patients with mitral regurgitation (MR) in the setting of coronary artery disease have a dismal long-term prognosis whether treated medically or surgically. Moreover, the optimal management of moderate ischemic MR at the time of coronary artery bypass grafting (CABG) remains the subjects of controversy. Thus, the present retrospective study was undertaken to determine whether mitral valve surgery for moderate ischemic MR at the time of CABG would be preferable to CABG alone in terms of clinical outcome. Material and Method: Between January 1997 and December 2003, 34 patients with moderate (Gr 3/4) ischemic MR underwent CABG alone (Group I, n=23) or CABG plus mitral valve surgery (Group II, n=11). Operative mortality, long-term survival and echocardiographic parameters were used to evaluate the efficacy of mitral valve surgery in patients with moderate ischemic MR. The mean follow-up durations of each group were $69.3\pm4.3$ months and $53.1\pm4.9$ months respectively. Result: There was no hospital mortality in both groups. There was one case of late mortality in Group I. The mean number of bypass graft was similar ($3.8\pm1.2\;vs\;3.7\pm1.3$ respectively). Cardiopulmonary bypass time was longer in group II (p=0.014). In group II, all of the patients received mitral annuloplasty using ring. On immediate postoperative echo-cardiogram, mitral regurgitation was reduced more in group II (p=0.002). Echocardiogram performed at last follow-up state showed no difference except the grade of MR between the two groups. Actuarial survival of both groups at 5 years was similar ($95.5\%\;vs\;100\%$, p=0.48). Conclusion: This study shows that in selected patients with moderate ischemic MR, CABG without mitral valve surgery might be sufficient. However, patients with low EF and NYHA functional class pre-operatively had tendency of significant residual MR, so mitral valve surgery should be necessary in these patients, and moreover, MR severity and left ventricle volume decreased more in mitral valve surgery group. Therefore, more large-scale studies are necessary to determine these effects on the ventricular function and long-term survival.

• KDI Journal of Economic Policy
• /
• v.32 no.4
• /
• pp.153-195
• /
• 2010

The recent global financial crisis has been the outcome of, among other things, the mismatch between institutions and the reality of the market in the current global financial system. The International financial institutions (IFIs) that were designed more than 60 years ago can no longer effectively meet the challenges posed by the current global economy. While the global financial market has become integrated like a single market, there is no international lender of last resort or global regulatory body. There also has been a rapid shift in the weight of economic power. The share of the Group of 7 (G7) countries in global gross domestic product (GDP) fell and the share of emerging market economies increased rapidly. Therefore, the tasks facing us today are: (i) to reform the IFIs -mandate, resources, management, and governance structure; (ii) to reform the system such as the international monetary system (IMS), and regulatory framework of the global financial system; and (iii) to reform global economic governance. The main focus of this paper will be the IMS reform and the role of the Group of Twenty (G20) summit meetings. The current IMS problems can be summarized as follows. First, the demand for foreign reserve accumulation has been increasing despite the movement from fixed exchange rate regimes to floating rate regimes some 40 years ago. Second, this increasing demand for foreign reserves has been concentrated in US dollar assets, especially public securities. Third, as the IMS relies too heavily on the supply of currency issued by a center country (the US), it gives an exorbitant privilege to this country, which can issue Treasury bills at the lowest possible interest rate in the international capital market. Fourth, as a related problem, the global financial system depends too heavily on the center country's ability to maintain the stability of the value of its currency and strength of its own financial system. Fifth, international capital flows have been distorted in the current IMS, from EMEs and developing countries where the productivity of capital investment is higher, to advanced economies, especially the US, where the return to capital investment is lower. Given these problems, there have been various proposals to reform the current IMS. They can be grouped into two: demand-side and supply-side reform. The key in the former is how to reduce the widespread strong demand for foreign reserve holdings among EMEs. There have been several proposals to reduce the self-insurance motivation. They include third-party insurance and the expansion of the opportunity to borrow from a global and regional reserve pool, or access to global lender of last resort (or something similar). However, the first option would be too costly. That leads us to the second option - building a stronger globalfinancial safety net. Discussions on supply-side reform of the IMS focus on how to diversify the supply of international reserve currency. The proposals include moving to a multiple currency system; increased allocation and wider use of special drawing rights (SDR); and creating a new global reserve currency. A key question is whether diversification should be encouraged among suitable existing currencies, or if it should be sought more with global reserve assets, acting as a complement or even substitute to existing ones. Each proposal has its pros and cons; they also face trade-offs between desirability and political feasibility. The transition would require close collaboration among the major players. This should include efforts at the least to strengthen policy coordination and collaboration among the major economies, and to reform the IMF to make it a more effective institution for bilateral and multilateral surveillance and as an international lender of last resort. The success on both fronts depends heavily on global economic governance reform and the role of the G20. The challenge is how to make the G20 effective. Without institutional innovations within the G20, there is a high risk that its summits will follow the path of previous summit meetings, such as G7/G8.

• Journal of the Mineralogical Society of Korea
• /
• v.19 no.4 s.50
• /
• pp.277-289
• /
• 2006

This study was carried out to find a new material having high removal efficiency for the harmful red tide. C. polykrikoides grow very fast and accumulate into dense and visible patches near the surface of the seawater ('Water bloom'). Some mineral medicines and Hwangto (reddish soil consist of clay minerals and Fe-oxides) were used in this study to remove C. polykrikoides. The pre-determined sprinkling ratio of mineral vs. seawater which contains approximately 5,000 cells/mL of C. polykrikoides was 10 g/L. In order to quantify the removal efficiency, the density of living cells was measured by counting with the Intervals of 0, 10, 30, and 60 minutes after sprinkling. Five Hwangtos feom different localities were examined in this study. It is found that a material with a high concentration of Fe and Al was the most effective to remove C. polykrikoides. After the sprinkling of the Hwangto showing the best removal efficiency in the test, 99% of total algaes were found to be eliminated within 60 minutes. Jeokeokji showed the highest removal efficiency among clay mineral medicines(92% removal efficiency after 60 minutes), and the rests in decreasing order are as follows: Gamto (91%) > Baekseokji (89%) > Hydromica (81%). In addition, Fe-oxide mineral medicine similarly looking as fine-grained earthy Daejaseok showed 100% removal efficiencyafter 30 minutes, and Wooyoeryang, 95% after 60 minutes. It is noted that even little addition (1 g/L) of Daejaseok, 10% of Hwangto concentration into seawater showed the removal efficiency of 100% after 60 minutes. From the results, it could be concluded that the fine-grained earthy Daejaseok was the most effective natural mineral medicine to remove the C. polykrikoides from seawater. Under the microscope the removal mechanism was found to be activated in the following order: adsorption, swelling of chain colony, chain colony crisis and algaecide.