• IEMEK Journal of Embedded Systems and Applications
• /
• v.15 no.4
• /
• pp.197-204
• /
• 2020

The ImageNet dataset is a large scale dataset and contains various natural scene images. In this paper, we propose a convolutional neural network (CNN)-based weighted ensemble technique for the ImageNet classification task. First, in order to fuse several models, our technique uses weights for each model, unlike the existing average-based ensemble technique. Then we propose an algorithm that automatically finds the coefficients used in later ensemble process. Our algorithm sequentially selects the model with the best performance of the validation set, and then obtains a weight that improves performance when combined with existing selected models. We applied the proposed algorithm to a total of 13 heterogeneous models, and as a result, 5 models were selected. These selected models were combined with weights, and we achieved 3.297% Top-5 error rate on the ImageNet test dataset.

• Journal of the Korean Society of Safety
• /
• v.31 no.2
• /
• pp.10-17
• /
• 2016

The purpose of this paper is to deduct components that are in the group of highest risk(top 10%). the group is conducted for classification into groups by values according to risk priority through risk priority number(RPN) of FMEA(Failure modes and effects analysis) sheet. Top 10% of failure mode among total potential failure modes(72 failure modes) of ESS included 5 BMS(battery included) failure modes, 1 invert failure mode, and 1 cable connectors failure mode in which BMS was highest. This is because ESS is connected to module, try, and lack in the battery part as an assembly of electronic information communication and is managed. BMS is mainly composed of the battery module and communication module. There is a junction box and numerous connectors that connect these two in which failure occurs most in the connector part and module itself. Finally, this paper proposes RPN by each step from the starting step of ESS design to installation and operation. Blackouts and electrical disasters can be prevented beforehand by managing and removing the deducted risk factors in prior.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• v.16 no.1
• /
• pp.44-51
• /
• 2016

Recently, artificial intelligence has reached the level of top information technologies that will have significant influence over many aspects of our future lifestyles. In particular, in the fields of machine learning technologies for classification and decision-making, there have been a lot of research efforts for solving estimation and control problems that appear in the various kinds of portfolio management problems via data-driven approaches. Note that these modern data-driven approaches, which try to find solutions to the problems based on relevant empirical data rather than mathematical analyses, are useful particularly in practical application domains. In this paper, we consider some applications of modern data-driven machine learning methods for portfolio management problems. More precisely, we apply a simplified version of the sparse Gaussian process (GP) classification method for classifying users' sensitivity with respect to financial risk, and then present two portfolio management issues in which the GP application results can be useful. Experimental results show that the GP applications work well in handling simulated data sets.

• Korean Journal of Adult Nursing
• /
• v.20 no.2
• /
• pp.296-310
• /
• 2008

Purpose: The purpose of this study was to compare the importance and performance of nursing interventions linked to five nursing diagnoses in CVA patients. Methods: First, total 37 nursing diagnoses were identified from the analysis of 78 nursing records of CVA patients, and then top 5 diagnoses were mapped with nursing interventions. Second, each intervention was compared in terms of importance and performance by 80 nurses working at neurosurgical units from 5 general hospitals. Data were analyzed using mean, SD, and t-test using the SPSS program. Results: Selected the top five nursing diagnoses were Acute Pain, Risk for Disuse Syndrome, Decreased Intracranial Adaptive Capacity, Ineffective Cerebral Tissue Perfusion and Acute Confusion. In general, most of the interventions were scored higher in importance than performance and most of independent interventions were not performed as frequently as it perceived in importance. The interventions which scored high in performance were the interventions ordered by physician or interventions related to medication behavior. Conclusion: We identified which nursing interventions should be performed more frequently and more critically important to nursing diagnoses. We recommend further research that enhances the performance of nursing interventions to provide better quality of nursing services to the patients in practice.

• Special Issue of the Society of Naval Architects of Korea
• /
• 2008.09a
• /
• pp.1-9
• /
• 2008

This report contains the results of structural analysis for the verification of the optimum offset range on the top of T-Bar with stiffener and BKT using at DSME Offset range as $6.0{\sim}10.0mm$ based on the 3-D FE analysis and experimental results of angie type stiffener as described in Annex 1 has been used as yard standard over ten (10) years under all Classification approval. Recently, Owner and Class have requested the confirmation for the misalignment based on the Yard's Standard so that a couple of locations for LNGC and LPGC has been investigated the structural strength by FE method using the offset ranges from 0.0 to 18.0 mm.

• Journal of KIISE
• /
• v.42 no.5
• /
• pp.558-565
• /
• 2015

Recently, the number of new malware and malware variants has dramatically increased. As a result, the time for analyzing malware and the efforts of malware analyzers have also increased. Therefore, malware classification helps malware analyzers decrease the overhead of malware analysis, and the classification is useful in studying the malware's genealogy. In this paper, we proposed a set of key opcode to classify the malware. In our experiments, we selected the top 10-opcode as key opcode, and the key opcode decreased the training time of a Supervised learning algorithm by 91% with preserving classification accuracy.

• Nuclear Engineering and Technology
• /
• v.43 no.2
• /
• pp.149-158
• /
• 2011

Metal waste generated from the pyroprocessing of 10 MtU of spent fuel was classified by comparing the specific activity of a relevant radionuclide with the limit value of the specific activity specified in the Korean acceptance criteria for a lowand intermediate-level waste repository. A Korean Optimized Fuel Assembly design with a 17${\times}$17 array, an initial enrichment of 4.5 weight-percent, discharge burn-up of 55 GWD/MtU, and a 10-year cooling time was considered. Initially, the mass and volume of each structural component of the assembly were calculated in detail, and a source term analysis was subsequently performed using ORIGEN-S for these components. An activation cross-section library generated by the KENO-VI/ORIGEN-S module was utilized for top-end and bottom-end pieces. As a result, an Inconel grid plate, a SUS plenum spring, a SUS guide tube subpart, SUS top-end and bottom-end pieces, and an Inconel top-end leaf spring were determined to be unacceptable for the Gyeongju low- and intermediate-level waste repository, as these waste products exceeded the acceptance criteria. In contrast, a Zircaloy grid plate and guide tube can be placed in the Gyeongju repository. Non-contaminated Zircaloy cladding occupying 76% of the metal waste was found to have a lower level of specific activity than the limit value. However, Zircaloy cladding contaminated by fission products and actinides during the decladding process of pyroprocessing was revealed to have 52 and 2 times higher specific activity levels than the limit values for alpha and $^{90}Sr$, respectively. Finally, it was found that 88.7% of the metal waste from the 17${\times}$17 Korean Optimized Fuel Assembly design should be disposed of in a deep geological repository. Therefore, it can be summarized that separation technology with a higher decontamination factor for transuranics and strontium should be developed for the efficient management of metal waste resulting from pyroprocessing.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.20 no.9
• /
• pp.1657-1665
• /
• 2016

Various fields are being researched through Deep Learning using CNN(Convolutional Neural Network) and these researches show excellent performance in the image recognition. In this paper, we provide streamlined GoogLeNet of CNN architecture that is capable of learning a large-scale Korean character database. The experimental data used in this paper is PHD08 that is the large-scale of Korean character database. PHD08 has 2,187 samples for each character and there are 2,350 Korean characters that make total 5,139,450 sample data. As a training result, streamlined GoogLeNet showed over 99% of test accuracy at PHD08. Also, we made additional Korean character data that have fonts that are not in the PHD08 in order to ensure objectivity and we compared the performance of classification between streamlined GoogLeNet and other OCR programs. While other OCR programs showed a classification success rate of 66.95% to 83.16%, streamlined GoogLeNet showed 89.14% of the classification success rate that is higher than other OCR program's rate.

• The Korean Journal of Ecology
• /
• v.23 no.6
• /
• pp.453-460
• /
• 2000

The forest vegetation of Mt. seondal was classified into eight communities and one afforestation by the phytosocialogical method (Z-M method). In general, Quercus mongolica trees occupied most of the area, while afforestation forest was distributed on the lower slope, cultivated land, and at the vicinity of village. The vegetation on the top part of Mt. Seondal was comparatively well preserved, but that in the lower areas has been disturbed heavily by human activity and some had mixed forests composed of pine trees, oaks, ashes, and Rhododendron micrantum shrub. By cluster analysis method. nine groups were identified as follows : Quercus mongolica group, Q. mongolica - Pinus densiflora group, Q. mongolica - Rhododendron schlipen - bachii group, Q. mongolica - Symplocos chinensis for. pilosa group, P. densiflora group, Juglans mandshurica group, Fraxinus mandshurica group, Betula costata group and Larix leptolepis group. These groups showed differences in species composition, but Quercus mongolica, Q. mongolica - P. densiflora, Q. mongolica - R. schlippenbachii and Q. mongolica - S. chinensis for. pilosa groups among them showed very similar floristic composition to each other. In the relationship between polar ordination axes and environmental variables, altitude was the environmental factor determining variation in species composition along axis X and soil moisture was the environmental along axis Y. They were the main factors in determining forest vegetation. The result of cluster analysis and polar ordination for the forest vegetation were corresponded to those of phytosocialogical classification in classifying vegetation.

• Korean Journal of Cognitive Science
• /
• v.26 no.3
• /
• pp.263-278
• /
• 2015

Online Q&A for the National Institute of the Korean Language provides expert's answers for questions about the Korean language, in which many similar questions are repeatedly posted like other Q&A boards. So, if a system automatically finds questions that are similar to a user's question, it can immediately provide users with recommendable answers to their question and prevent experts from wasting time to answer to similar questions repeatedly. In this paper, we set 5 classes of questions based on its topic which are frequently asked, and propose to classify questions to those classes. Our system searches similar questions by combining topic similarity, vector similarity and sequence similarity. Experiment shows that our method improves search correctness with topic classification. In experiment, Mean Reciprocal Rank(MRR) of our system is 0.756, and precision for the first result is 68.31% and precision for top five results is 87.32%.