• Journal of Intelligence and Information Systems
• /
• v.16 no.4
• /
• pp.99-112
• /
• 2010

Ensemble learning is a method for improving the performance of classification and prediction algorithms. It is a method for finding a highly accurateclassifier on the training set by constructing and combining an ensemble of weak classifiers, each of which needs only to be moderately accurate on the training set. Ensemble learning has received considerable attention from machine learning and artificial intelligence fields because of its remarkable performance improvement and flexible integration with the traditional learning algorithms such as decision tree (DT), neural networks (NN), and SVM, etc. In those researches, all of DT ensemble studies have demonstrated impressive improvements in the generalization behavior of DT, while NN and SVM ensemble studies have not shown remarkable performance as shown in DT ensembles. Recently, several works have reported that the performance of ensemble can be degraded where multiple classifiers of an ensemble are highly correlated with, and thereby result in multicollinearity problem, which leads to performance degradation of the ensemble. They have also proposed the differentiated learning strategies to cope with performance degradation problem. Hansen and Salamon (1990) insisted that it is necessary and sufficient for the performance enhancement of an ensemble that the ensemble should contain diverse classifiers. Breiman (1996) explored that ensemble learning can increase the performance of unstable learning algorithms, but does not show remarkable performance improvement on stable learning algorithms. Unstable learning algorithms such as decision tree learners are sensitive to the change of the training data, and thus small changes in the training data can yield large changes in the generated classifiers. Therefore, ensemble with unstable learning algorithms can guarantee some diversity among the classifiers. To the contrary, stable learning algorithms such as NN and SVM generate similar classifiers in spite of small changes of the training data, and thus the correlation among the resulting classifiers is very high. This high correlation results in multicollinearity problem, which leads to performance degradation of the ensemble. Kim,s work (2009) showedthe performance comparison in bankruptcy prediction on Korea firms using tradition prediction algorithms such as NN, DT, and SVM. It reports that stable learning algorithms such as NN and SVM have higher predictability than the unstable DT. Meanwhile, with respect to their ensemble learning, DT ensemble shows the more improved performance than NN and SVM ensemble. Further analysis with variance inflation factor (VIF) analysis empirically proves that performance degradation of ensemble is due to multicollinearity problem. It also proposes that optimization of ensemble is needed to cope with such a problem. This paper proposes a hybrid system for coverage optimization of NN ensemble (CO-NN) in order to improve the performance of NN ensemble. Coverage optimization is a technique of choosing a sub-ensemble from an original ensemble to guarantee the diversity of classifiers in coverage optimization process. CO-NN uses GA which has been widely used for various optimization problems to deal with the coverage optimization problem. The GA chromosomes for the coverage optimization are encoded into binary strings, each bit of which indicates individual classifier. The fitness function is defined as maximization of error reduction and a constraint of variance inflation factor (VIF), which is one of the generally used methods to measure multicollinearity, is added to insure the diversity of classifiers by removing high correlation among the classifiers. We use Microsoft Excel and the GAs software package called Evolver. Experiments on company failure prediction have shown that CO-NN is effectively applied in the stable performance enhancement of NNensembles through the choice of classifiers by considering the correlations of the ensemble. The classifiers which have the potential multicollinearity problem are removed by the coverage optimization process of CO-NN and thereby CO-NN has shown higher performance than a single NN classifier and NN ensemble at 1% significance level, and DT ensemble at 5% significance level. However, there remain further research issues. First, decision optimization process to find optimal combination function should be considered in further research. Secondly, various learning strategies to deal with data noise should be introduced in more advanced further researches in the future.

• Journal of Intelligence and Information Systems
• /
• v.16 no.2
• /
• pp.129-141
• /
• 2010

An Internet shopping mall for clothing operates a warehouse for packing and shipping products to fulfill its orders. All the products in the warehouse are put into the boxes of same brands and the boxes are stored in a row on shelves equiped in the warehouse. To make picking and managing easy, boxes of the same brands are located side by side on the shelves. When new products arrive to the warehouse for storage, the products of a brand are put into boxes and those boxes are located adjacent to the boxes of the same brand. If there is not enough space for the new coming boxes, however, some boxes of other brands should be moved away and then the new coming boxes are located adjacent in the resultant vacant spaces. We want to minimize the movement of the existing boxes of other brands to another places on the shelves during the warehousing of new coming boxes, while all the boxes of the same brand are kept side by side on the shelves. Firstly, we define the adjacency of boxes by looking the shelves as an one dimensional series of spaces to store boxes, i.e. cells, tagging the series of cells by a series of numbers starting from one, and considering any two boxes stored in the cells to be adjacent to each other if their cell numbers are continuous from one number to the other number. After that, we tried to formulate the problem into an integer programming model to obtain an optimal solution. An integer programming formulation and Branch-and-Bound technique for this problem may not be tractable because it would take too long time to solve the problem considering the number of the cells or boxes in the warehouse and the computing power of the Internet shopping mall. As an alternative approach, we designed a fast heuristic method for this reallocation problem by focusing on just the unused spaces-empty cells-on the shelves, which results in an assignment problem model. In this approach, the new coming boxes are assigned to each empty cells and then those boxes are reorganized so that the boxes of a brand are adjacent to each other. The objective of this new approach is to minimize the movement of the boxes during the reorganization process while keeping the boxes of a brand adjacent to each other. The approach, however, does not ensure the optimality of the solution in terms of the original problem, that is, the problem to minimize the movement of existing boxes while keeping boxes of the same brands adjacent to each other. Even though this heuristic method may produce a suboptimal solution, we could obtain a satisfactory solution within a satisfactory time, which are acceptable by real world experts. In order to justify the quality of the solution by the heuristic approach, we generate 100 problems randomly, in which the number of cells spans from 2,000 to 4,000, solve the problems by both of our heuristic approach and the original integer programming approach using a commercial optimization software package, and then compare the heuristic solutions with their corresponding optimal solutions in terms of solution time and the number of movement of boxes. We also implement our heuristic approach into a storage location assignment system for the Internet shopping mall.

• Korean Journal of Veterinary Research
• /
• v.10 no.1
• /
• pp.11-23
• /
• 1970

Recently Carter(1952) reported the capsule antigens of Pasteurella multocida could be divided into four serological types A,B,C and D by means of precipitation tests. Subsequently he showed that the most sensitive for identification of these types involved the use of capsule substance adsorbed by erythrocytes in hemagglutination test. It may be somewhat difficult to conduct the hemagglutination test in small laboratory, because relatively large amounts of antisera and erythrocytes of the human O type are required for the test. A simple method for serological typing of P. multocida was the slide agglutination test employed by Little et al. (1943) and Namioka et al. (1962), but this method is still in controversy. The author tried adapting Carter's hemagglutination method to the slide method so called "micromethod technique", and studied on the stabilization of erythrocytes for use of slide hemagglutination to P. multocida although many invesigators reported the stabilization of erythrocytes. The results obtained are summarized as follows: 1. A simplified method (slide method) for capsule typing of the organism was developed by adapting Carter's hemagglutination reaction(tube method). Antibody-containing serum can be diluted serially on Boerner's microtest slide with capillary or serological pipetts with a considerable accuracy. The slide reaction can be carried out with case on the slide by adding $0.05m{\ell}$ of antigen-sensitized erythrocytes suspension diluted to one percent on $0.05m{\ell}$ of serially diluted antibody-containing sera, and the final result can be read after 60 minutes at the room temperature ($15^{\circ}C$). 2. It is difficult to determine superiority of inferiority between the slide method and the tube method on the pattern of the reaction of hemagglutination. 3. The pH range of 6.6 to 8.3 is optimal for the slide hemagglutination reaction. 4. The antigen-sensitization against erythrocytes at $37^{\circ}C$ is optimal for the slide hemagglutination. 5. Both the doses and concentration of antigen do not influence the antigen-adsorbing capacity of erythrocytes. 6. The reduction of antigen-sensitizing hours does not influence the antigen-adsorbing capacity of erythrocytes even 30 minutes. 7. The tannic acid treatment against formalinized and non-formalinized erythrocytes showed no effect on the reaction of hemagglutination. 8. The erythrocytes preserved at $4^{\circ}C$ in the ACD solution do not decrease the reactivity on the reaction of hemagglutination for 60 days, while they begin slight hemolysis 30 days after preserving. 9. The stable preparation of erythrocytes can be obtained by treating the cells at $37^{\circ}C$ for 20 hours with from 4 to 8 percent of formalin in saline or buffer. These cells can be preserved at $4^{\circ}C$ for more than 8 months experimented without hemolysis. With low concentration of formalin, the cells were not sufficiently stabilized resulting in the hemolysis after short period of preservation at $4^{\circ}C$. 10. The erythrocytes treated with 16 percent of formalin remain constantly or increase the reactivity for the reaction of hemagglutination. On the contrary, the cells treated with I to 8 percent of formalin decrease the reactivity. 11. There is no difference between nontreated fresh erythrocytes and the erythrocytes preserved in the ACD solution on the reactivity against the hemagglutination, and the erythrocytes treated with 16 percent of formalin showed the reactivity of higher level than that of the above two kinds of erythrocytes. 12. There is no difference between the saline and the isotonic buffer solution on the reaction of hemagglutination.

• Journal of Korean Society of Disaster and Security
• /
• v.15 no.1
• /
• pp.1-12
• /
• 2022

An automatic optimization technique is used to estimate the optimal parameters of the hydrologic model, and different hydrologic response results can be provided depending on objective functions. In this study, the parameters of the event-based rainfall-runoff model were estimated using various objective functions, the reproducibility of the hydrograph according to the objective functions was evaluated, and appropriate objective functions were proposed. As the rainfall-runoff model, the storage function model(SFM), which is a lumped hydrologic model used for runoff simulation in the current Korean flood forecasting system, was selected. In order to evaluate the reproducibility of the hydrograph for each objective function, 9 rainfall events were selected for the Cheoncheon basin, which is the upstream basin of Yongdam Dam, and widely-used 7 objective functions were selected for parameter estimation of the SFM for each rainfall event. Then, the reproducibility of the simulated hydrograph using the optimal parameter sets based on the different objective functions was analyzed. As a result, RMSE, NSE, and RSR, which include the error square term in the objective function, showed the highest accuracy for all rainfall events except for Event 7. In addition, in the case of PBIAS and VE, which include an error term compared to the observed flow, it also showed relatively stable reproducibility of the hydrograph. However, in the case of MIA, which adjusts parameters sensitive to high flow and low flow simultaneously, the hydrograph reproducibility performance was found to be very low.

• KIPS Transactions on Computer and Communication Systems
• /
• v.10 no.8
• /
• pp.223-230
• /
• 2021

The lightweight block cipher PIPO announced at ICISC'20 has been effectively implemented by applying the bit slice technique. In this paper, we propose a parallel optimal implementation of PIPO for ARM processors. The proposed implementation enables parallel encryption of 8-plaintexts and 16-plaintexts. The implementation targets the A10x fusion processor. On the target processor, the existing reference PIPO code has performance of 34.6 cpb and 44.7 cpb in 64/128 and 64/256 standards. Among the proposed methods, the general implementation has a performance of 12.0 cpb and 15.6 cpb in the 8-plaintexts 64/128 and 64/256 standards, and 6.3 cpb and 8.1 cpb in the 16-plaintexts 64/128 and 64/256 standards. Compared to the existing reference code implementation, the 8-plaintexts parallel implementation for each standard has about 65.3%, 66.4%, and the 16-plaintexts parallel implementation, about 81.8%, and 82.1% better performance. The register minimum alignment implementation shows performance of 8.2 cpb and 10.2 cpb in the 8-plaintexts 64/128 and 64/256 specifications, and 3.9 cpb and 4.8 cpb in the 16-plaintexts 64/128 and 64/256 specifications. Compared to the existing reference code implementation, the 8-plaintexts parallel implementation has improved performance by about 76.3% and 77.2%, and the 16-plaintext parallel implementation is about 88.7% and 89.3% higher for each standard.

• Journal of KIISE:Computer Systems and Theory
• /
• v.34 no.8
• /
• pp.367-376
• /
• 2007

In this paper, we introduce an application-specific and adaptive power management technique for portable systems that support dynamic voltage scaling (DVS). We exploit both the idle time of multitasking systems running soft real-time tasks as well as memory- or CPU-bound code regions. Detailed power and execution time profiles guide an adaptive power manager (APM) that is linked to the operating system. A post-pass optimizer marks candidate regions for DVS by inserting calls to the APM. At runtime, the APM monitors the CPU's performance counters to dynamically determine the affinity of the each marked region. for each region, the APM computes the optimal voltage and frequency setting in terms of energy consumption and switches the CPU to that setting during the execution of the region. Idle time is exploited by monitoring system idle time and switching to the energy-wise most economical setting without prolonging execution. We show that our method is most effective for periodic workloads such as video or audio decoding. We have implemented our method in a multitasking operating system (Microsoft Windows CE) running on an Intel XScale-processor. We achieved up to 9% of total system power savings over the standard power management policy that puts the CPU in a low Power mode during idle periods.

• Journal of Yeungnam Medical Science
• /
• v.4 no.2
• /
• pp.113-120
• /
• 1987

The effect of intravesical formalin instillation as a therapeutic modality for intractable bladder hemorrhage is well known. And despite clear evidence of therapeutic efficacy of intravesical cytotoxic drugs and/or BCG immunotherapy, there have been substantial recurrences during followup after transurethral resection for superficial bladder tumor. If formalin injected at the bed of superficial bladder tumor is able to coagulate and necrotize the tumor, it will be greatly helpful to the patients With recurrent bladder tumor developed during followup. Since this technique is applicable on outpatient basis, an economical as well as a psychological burden of the patients can be reduced considerably. The purpose of this study is to evaluate the effect of submucosal formalin injection on rat bladder wall, 36 healthy adult male Sprague-Dawley rats (weighing 350gm in average) were divided into 3 groups: In Group I (control group), 0.01ml of normal saline was injected submucosally at the left posterolateral wall of the bladder opened under intraperitoneal Nembutal anesthesia ; In Group II and III, 0.01 ml of 10% and 4% formalin, respectively, were administered at the same site as in the Group I, two rats in each group were sacrificed at day 1, 2, and 3, and week 1, 2 and 4 after injection, respectively. Gross and microscopic examination of the cystectomized specimen were done in each group. In the Group II, bladder stones were formed at week I, and in both the Group I and III, stones were seen at week 2 post injection. There was no significant difference III histologic findings of the bladder between the group II and III. Mucosal ulcer and/or prominent mucosal disruption was observed at 24 hours after injection in both Group II and III. Epithelial regeneration began at day 2, and was marked at day 3, and epithelial lining was almost normalized one week after injection. Subepithelial edema, telangiectasia and inflammatory reaction were prominent at 24 hours post formalin injection. Subepithelial edema persisted in moderate degree for 1 week. Telangiectasia and inflammatory reaction were noted for 4 weeks. Mild degree of these findings also appeared In the control group. Fibroblastic proliferation appeared at day 2 and persisted in moderate degree for 4 weeks. There has been no mortality or bladder perforation. These results suggest that clinical application of this technique is feasible for the selected cases of recurrent, solitary superficial bladder tumor. However, optimal dosage of formalin in relation to the size of the lesion remains to be investigated.

• Radiation Oncology Journal
• /
• v.20 no.1
• /
• pp.1-16
• /
• 2002

Purpose : Three dimensional conformal radiotherapy planning is being used widely for the treatment of patients with brain tumor. However, it takes much time to develop an optimal treatment plan, therefore, it is difficult to apply this technique to all patients. To increase the efficiency of this technique, we need to develop standard radiotherapy plant for each site of the brain. Therefore we developed several 3 dimensional conformal radiotherapy plans (3D plans) for tumors at each site of brain, compared them with each other, and with 2 dimensional radiotherapy plans. Finally model plans for each site of the brain were decide. Materials and Methods : Imaginary tumors, with sizes commonly observed in the clinic, were designed for each site of the brain and drawn on CT images. The planning target volumes (PTVs) were as follows; temporal $tumor-5.7\times8.2\times7.6\;cm$, suprasellar $tumor-3\times4\times4.1\;cm$, thalamic $tumor-3.1\times5.9\times3.7\;cm$, frontoparietal $tumor-5.5\times7\times5.5\;cm$, and occipitoparietal $tumor-5\times5.5\times5\;cm$. Plans using paralled opposed 2 portals and/or 3 portals including fronto-vertex and 2 lateral fields were developed manually as the conventional 2D plans, and 3D noncoplanar conformal plans were developed using beam's eye view and the automatic block drawing tool. Total tumor dose was 54 Gy for a suprasellar tumor, 59.4 Gy and 72 Gy for the other tumors. All dose plans (including 2D plans) were calculated using 3D plan software. Developed plans were compared with each other using dose-volume histograms (DVH), normal tissue complication probabilities (NTCP) and variable dose statistic values (minimum, maximum and mean dose, D5, V83, V85 and V95). Finally a best radiotherapy plan for each site of brain was selected. Results : 1) Temporal tumor; NTCPs and DVHs of the normal tissue of all 3D plans were superior to 2D plans and this trend was more definite when total dose was escalated to 72 Gy (NTCPs of normal brain 2D $plans:27\%,\;8\%\rightarrow\;3D\;plans:1\%,\;1\%$). Various dose statistic values did not show any consistent trend. A 3D plan using 3 noncoplanar portals was selected as a model radiotherapy plan. 2) Suprasellar tumor; NTCPs of all 3D plans and 2D plans did not show significant difference because the total dose of this tumor was only 54 Gy. DVHs of normal brain and brainstem were significantly different for different plans. D5, V85, V95 and mean values showed some consistent trend that was compatible with DVH. All 3D plans were superior to 2D plans even when 3 portals (fronto-vertex and 2 lateral fields) were used for 2D plans. A 3D plan using 7 portals was worse than plans using fewer portals. A 3D plan using 5 noncoplanar portals was selected as a model plan. 3) Thalamic tumor; NTCPs of all 3D plans were lower than the 2D plans when the total dose was elevated to 72 Gy. DVHs of normal tissues showed similar results. V83, V85, V95 showed some consistent differences between plans but not between 3D plans. 3D plans using 5 noncoplanar portals were selected as a model plan. 4) Parietal (fronto- and occipito-) tumors; all NTCPs of the normal brain in 3D plans were lower than in 2D plans. DVH also showed the same results. V83, V85, V95 showed consistent trends with NTCP and DVH. 3D plans using 5 portals for frontoparietal tumor and 6 portals for occipitoparietal tumor were selected as model plans. Conclusion : NTCP and DVH showed reasonable differences between plans and were through to be useful for comparing plans. All 3D plans were superior to 2D plans. Best 3D plans were selected for tumors in each site of brain using NTCP, DVH and finally by the planner's decision.

• Korean Journal of Agricultural and Forest Meteorology
• /
• v.15 no.1
• /
• pp.1-8
• /
• 2013

This study was conducted to compare the local climate conditions and growth periods for the apple (Malus domestica 'Borkh') orchards at different altitudes (330, 500, and 670 m) in Jangsu-Gun, Korea. Observation data for the growth period show that the monthly mean air temperatures at the 'Hongro' apple orchard sites decrease with height at the rate of 1.0 to $3.0^{\circ}C$/100 m. The monthly minimum temperatures in April (blooming period for 'Hongro' apple) were 4.3, 2.9, and $0.4^{\circ}C$ at 330, 500, and 670 m, respectively. The monthly mean temperatures in September (i.e., the coloration and maturation period) were 20.6, 18.7, and $14.5^{\circ}C$, respectively. The annual precipitation range varied from 1,234 to 1,439 mm, which tended to increase with height. The heavy rainfall occurred in summer (June to August) and amounted to 827-933 mm. No significant differences in the duration of sunshine were observed amongst the orchards at three different altitudes. The earliest bud break was observed at the 330 m altitude (18 March 2009), which was 4 and 11 days earlier in comparison to those at 500 and 670 m, respectively. The time of full bloom at 330 m was 12 days ahead of that at 670 m. The optimal maturation of fruit (based on skin redness > 80%) was observed between 7 and 10 September at 330 m, 15 and 18 September at 500 m, and 21 and 23 September at 670 m.

• Journal of Bio-Environment Control
• /
• v.25 no.4
• /
• pp.232-239
• /
• 2016

The objective of this study was to examine the effect of photoperiod and light intensity of RBW LED (red:blue:white = 2:1:1) on the growth of Taraxacum coreanum Nakai in a fully artificial light type plant factory. 3 photoperiods and 4 light intensity were used respectively in a fully artificial light type plant production system. Plants were cultured with three photoperiods and four light intensity regimes (conditions) for 270 and 120 days, respectively, using nutrient film technique (NFT) or aeroponics culture methods. For each photoperiod, the total leaves per plant harvested 8 times in all cultivation period was 224 in the 16/8(day/light) photoperiod that had no significant difference from 220 in the 12/12 photoperiod and the lowest number of leaves was 151 occurred in the 8/16 photoperiod, which means that the longer photoperiod, the more leaves harvest. Total fresh weight of above ground was the high in order of in 16/8 photoperiod as 125g, 12/12 photoperiod as 91g, 8/16 photoperiod as 56g. For each light intensity, the total leaves per plant harvested 4 times in all cultivation period was the great in order of $150{\mu}mol{\cdot}m^{-2}{\cdot}s^{-1}$ as 123, $200{\mu}mol{\cdot}m^{-2}{\cdot}s^{-1}$ as 107, $100{\mu}mol{\cdot}m^{-2}{\cdot}s^{-1}$ as 95, $50{\mu}mol{\cdot}m^{-2}{\cdot}s^{-1}$ as 56 which was the smallest number of total leaves harvest. Total fresh weight of above ground per plant was the high in order of $150{\mu}mol{\cdot}m^{-2}{\cdot}s^{-1}$ as 43.6g, $100{\mu}mol{\cdot}m^{-2}{\cdot}s^{-1}$ as 34.6g, $200{\mu}mol{\cdot}m^{-2}{\cdot}s^{-1}$ as 32.2g, $50{\mu}mol{\cdot}m^{-2}{\cdot}s^{-1}$ as 18.2g. From these results, it was concluded that photoperiod of 16/8 and light intensity at $150{\mu}mol{\cdot}m^{-2}{\cdot}s^{-1}$ can be used as the light condition of RBW LED (red:blue:white = 2:1:1) for optimal growth of Taraxacum coreanum Nakai in a fully artificial light type plant factory.