• Journal of Intelligence and Information Systems
• /
• v.24 no.1
• /
• pp.205-225
• /
• 2018

Convolutional Neural Network (ConvNet) is one class of the powerful Deep Neural Network that can analyze and learn hierarchies of visual features. Originally, first neural network (Neocognitron) was introduced in the 80s. At that time, the neural network was not broadly used in both industry and academic field by cause of large-scale dataset shortage and low computational power. However, after a few decades later in 2012, Krizhevsky made a breakthrough on ILSVRC-12 visual recognition competition using Convolutional Neural Network. That breakthrough revived people interest in the neural network. The success of Convolutional Neural Network is achieved with two main factors. First of them is the emergence of advanced hardware (GPUs) for sufficient parallel computation. Second is the availability of large-scale datasets such as ImageNet (ILSVRC) dataset for training. Unfortunately, many new domains are bottlenecked by these factors. For most domains, it is difficult and requires lots of effort to gather large-scale dataset to train a ConvNet. Moreover, even if we have a large-scale dataset, training ConvNet from scratch is required expensive resource and time-consuming. These two obstacles can be solved by using transfer learning. Transfer learning is a method for transferring the knowledge from a source domain to new domain. There are two major Transfer learning cases. First one is ConvNet as fixed feature extractor, and the second one is Fine-tune the ConvNet on a new dataset. In the first case, using pre-trained ConvNet (such as on ImageNet) to compute feed-forward activations of the image into the ConvNet and extract activation features from specific layers. In the second case, replacing and retraining the ConvNet classifier on the new dataset, then fine-tune the weights of the pre-trained network with the backpropagation. In this paper, we focus on using multiple ConvNet layers as a fixed feature extractor only. However, applying features with high dimensional complexity that is directly extracted from multiple ConvNet layers is still a challenging problem. We observe that features extracted from multiple ConvNet layers address the different characteristics of the image which means better representation could be obtained by finding the optimal combination of multiple ConvNet layers. Based on that observation, we propose to employ multiple ConvNet layer representations for transfer learning instead of a single ConvNet layer representation. Overall, our primary pipeline has three steps. Firstly, images from target task are given as input to ConvNet, then that image will be feed-forwarded into pre-trained AlexNet, and the activation features from three fully connected convolutional layers are extracted. Secondly, activation features of three ConvNet layers are concatenated to obtain multiple ConvNet layers representation because it will gain more information about an image. When three fully connected layer features concatenated, the occurring image representation would have 9192 (4096+4096+1000) dimension features. However, features extracted from multiple ConvNet layers are redundant and noisy since they are extracted from the same ConvNet. Thus, a third step, we will use Principal Component Analysis (PCA) to select salient features before the training phase. When salient features are obtained, the classifier can classify image more accurately, and the performance of transfer learning can be improved. To evaluate proposed method, experiments are conducted in three standard datasets (Caltech-256, VOC07, and SUN397) to compare multiple ConvNet layer representations against single ConvNet layer representation by using PCA for feature selection and dimension reduction. Our experiments demonstrated the importance of feature selection for multiple ConvNet layer representation. Moreover, our proposed approach achieved 75.6% accuracy compared to 73.9% accuracy achieved by FC7 layer on the Caltech-256 dataset, 73.1% accuracy compared to 69.2% accuracy achieved by FC8 layer on the VOC07 dataset, 52.2% accuracy compared to 48.7% accuracy achieved by FC7 layer on the SUN397 dataset. We also showed that our proposed approach achieved superior performance, 2.8%, 2.1% and 3.1% accuracy improvement on Caltech-256, VOC07, and SUN397 dataset respectively compare to existing work.

• The Journal of Engineering Geology
• /
• v.21 no.2
• /
• pp.133-145
• /
• 2011

Unsaturated soil column tests were performed for weathered gneiss soil and weathered granite soil to assess the relationship between infiltration velocity and rainfall condition for different rainfall durations and for multiple rainfall events separated by dry periods of various lengths (herein, 'rainfall break duration'). The volumetric water content was measured using TDR (Time Domain Reflectometry) sensors at regular time intervals. For the column tests, rainfall intensity was 20 mm/h and we varied the rainfall duration and rainfall break duration. The unit weight of weathered gneiss soil was designed 1.21 $g/cm^3$, which is lower than the in situ unit weight without overflow in the column. The in situ unit weight for weathered granite soil was designed 1.35 $g/cm^3$. The initial infiltration velocity of precipitation for the two weathered soils under total amount of rainfall as much as 200 mm conditions was $2.090{\times}10^{-3}$ to $2.854{\times}10^{-3}$ cm/s and $1.692{\times}10^{-3}$ to $2.012{\times}10^{-3}$ cm/s, respectively. These rates are higher than the repeated-infiltration velocities of precipitation under total amount of rainfall as much as 100 mm conditions ($1.309{\times}10^{-3}$ to $1.871{\times}10^{-3}$ cm/s and $1.175{\times}10^{-3}$ to $1.581{\times}10^{-3}$ cm/s, respectively), because the amount of precipitation under 200 mm conditions is more than that under 100 mm conditions. The repeated-infiltration velocities of weathered gneiss soil and weathered granite soil were $1.309{\times}10^{-3}$ to $2.854{\times}10^{-3}$ cm/s and $1.175{\times}10^{-3}$ to $2.012{\times}10^{-3}$ cm/s, respectively, being higher than the first-infiltration velocities ($1.307{\times}10^{-2}$ to $1.718{\times}10^{-2}$ cm/s and $1.789{\times}10^{-2}$ to $2.070{\times}10^{-2}$ cm/s, respectively). The results reflect the effect of reduced matric suction due to a reduction in the amount of air in the soil.

• 한국균학회소식:학술대회논문집
• /
• 2015.05a
• /
• pp.59-59
• /
• 2015

Cryptococcus neoformans causes life-threatening meningoencephalitis in humans, but the treatment of cryptococcosis remains challenging. To develop novel therapeutic targets and approaches, signaling cascades controlling pathogenicity of C. neoformans have been extensively studied but the underlying biological regulatory circuits remain elusive, particularly due to the presence of an evolutionarily divergent set of transcription factors (TFs) in this basidiomycetous fungus. In this study, we constructed a high-quality of 322 signature-tagged gene deletion strains for 155 putative TF genes, which were previously predicted using the DNA-binding domain TF database (http://www.transcriptionfactor.org/). We tested in vivo and in vitro phenotypic traits under 32 distinct growth conditions using 322 TF gene deletion strains. At least one phenotypic trait was exhibited by 145 out of 155 TF mutants (93%) and approximately 85% of the TFs (132/155) have been functionally characterized for the first time in this study. Through high-coverage phenome analysis, we discovered myriad novel TFs that play critical roles in growth, differentiation, virulence-factor (melanin, capsule, and urease) formation, stress responses, antifungal drug resistance, and virulence. Large-scale virulence and infectivity assays in insect (Galleria mellonella) and mouse host models identified 34 novel TFs that are critical for pathogenicity. The genotypic and phenotypic data for each TF are available in the C. neoformans TF phenome database (http://tf.cryptococcus.org). In conclusion, our phenome-based functional analysis of the C. neoformans TF mutant library provides key insights into transcriptional networks of basidiomycetous fungi and ubiquitous human fungal pathogens.

• The Korean Journal of Air & Space Law and Policy
• /
• v.35 no.2
• /
• pp.75-111
• /
• 2020

The civilian drone world has evolved in recent years from one dominated by hobbyists to growing involvement by companies seeking to profit from unmanned flight in everything from infrastructure inspections to drone deliveries that are already subject to regulations. Drone flight under the property right relation with the land owner would be deemed legal on the condition that expeditious and innocent passage of drone flight over the land be assured. The United Nations Convention on the Law of the Sea (UNCLOS) enshrines the concept of innocent passage through a coastal state's territorial sea. Passage is innocent so long as it is not prejudicial to the peace, good order or security of the coastal state. A vessel in innocent passage may traverse the coastal state's territorial sea continuously and expeditiously, not stopping or anchoring except in force majeure situations. However, the disturbances caused by drone flight may be removed, which is defined as infringement against the constitutional interest of personal rights. For example, aggressive infringement against privacy and personal freedom may be committed by drone more easily than ever before, and than other means. The cost-benefit analysis, however, has been recognjzed as effective criteria regarding the removal of disturbances or injunction decision. Applying that analysis, the civil action against such infringement may not find suitable basis for making a good case. Because the removal of such infringement through civil actions may result in only the deletion of journal article. The injunction of drone flight before taking the information would not be obtainable through civil action, Therefore, more detailed and meticulous regulation and criteria in public law domain may be preferable than civil action, at present time. It may be suitable for legal stability and drone industry to set up the detailed public regulations restricting the free flight of drone capable of acquiring visual information amounting to the infrigement against the right of personal information security.

• The Korean Journal of Nuclear Medicine Technology
• /
• v.14 no.1
• /
• pp.46-53
• /
• 2010

Purpose: The static image of nuclear medicine study should be acquired without a motion, however, it is difficult to acquire static image without movement for the serious patients, advanced aged patients. These movements cause decreases in reliability for quantitative and qualitative analysis, therefore re-examination was inevitable in the some cases. Consequently, in order to improve the problem of motion artifacts, the authors substituted the dynamic acquisition technique for the static acquisition, using motion correction. Materials and Methods: A capillary tube and IEC body phantom were used. First, the static image was acquired for 60 seconds while the dynamic images were acquired with a protocol, 2 sec/frame${\times}$30 frames, under the same parameter and the frames were summed up into one image afterwards. Also, minimal motion and excessive motion were applied during the another dynamic acquisition and the coordinate correction was applied towards X and Y axis on the frames where the motion artifact occurred. But the severe blurred images were deleted. Finally, the resolution and counts were compared between the static image and the summed dynamic images which before and after applying motion correction, and the signal of frequency was analysed after frequency spatial domain was transformed into 2D FFT. Supplementary examination, the blind test was performed by the nuclear medicine department staff. Results: First, the resolution in the static image and summed dynamic image without motion were 8.32 mm, 8.37 mm on X-axis and 8.30 mm, 8.42 mm on Y-axis, respectively. The counts were 484 kcounts, 485 kcounts each, so there was nearly no difference. Secondly, the resolution in the image with minimal motion applying motion correction was 8.66 mm on X-axis, 8.85 mm on Y-axis and had 469 kcounts while the image without motion correction was 21.81 mm, 24.02 mm and 469 kcounts in order. So, this shows the image with minimal motion applying motion correction has similar resolution with the static image. Lastly, the resolution in the images with excessive motion applying motion correction were 9.09 mm on X-axis, 8.83 mm on Y-axis and had 469 kcounts while the image without motion correction was 47.35 mm, 40.46 mm and 255 kcounts in order. Although there was difference in counts because of deletion of blurred frames, we could get similar resolution. And when the image was transformed into frequency, the high frequency was decreased by the movement. However, the frequency was improved again after motion correction. In the blind test, there was no difference between the image applying motion correction and the static image without motion. Conclusion: There was no significant difference between the static image and the summed dynamic image. This technique can be applied to patients who may have difficulty remaining still during the imaging process, so that the quality of image can be improved as well as the reliance for analysis of quantity. Moreover, the re-examination rate will be considerably decreased. However, there is a limit of motion correction, more time will be required to successfully image the patients applying motion correction. Also, the decrease of total counts due to deletion of the severe blurred images should be calculated and the proper number of frames should be acquired.

• Asia pacific journal of information systems
• /
• v.19 no.2
• /
• pp.139-155
• /
• 2009

For the last few decades, many studies have tried to explore and unveil venture companies' success factors and unique features in order to identify the sources of such companies' competitive advantages over their rivals. Such venture companies have shown tendency to give high returns for investors generally making the best use of information technology. For this reason, many venture companies are keen on attracting avid investors' attention. Investors generally make their investment decisions by carefully examining the evaluation criteria of the alternatives. To them, credit rating information provided by international rating agencies, such as Standard and Poor's, Moody's and Fitch is crucial source as to such pivotal concerns as companies stability, growth, and risk status. But these types of information are generated only for the companies issuing corporate bonds, not venture companies. Therefore, this study proposes a method for evaluating venture businesses by presenting our recent empirical results using financial data of Korean venture companies listed on KOSDAQ in Korea exchange. In addition, this paper used multi-class SVM for the prediction of DEA-based efficiency rating for venture businesses, which was derived from our proposed method. Our approach sheds light on ways to locate efficient companies generating high level of profits. Above all, in determining effective ways to evaluate a venture firm's efficiency, it is important to understand the major contributing factors of such efficiency. Therefore, this paper is constructed on the basis of following two ideas to classify which companies are more efficient venture companies: i) making DEA based multi-class rating for sample companies and ii) developing multi-class SVM-based efficiency prediction model for classifying all companies. First, the Data Envelopment Analysis(DEA) is a non-parametric multiple input-output efficiency technique that measures the relative efficiency of decision making units(DMUs) using a linear programming based model. It is non-parametric because it requires no assumption on the shape or parameters of the underlying production function. DEA has been already widely applied for evaluating the relative efficiency of DMUs. Recently, a number of DEA based studies have evaluated the efficiency of various types of companies, such as internet companies and venture companies. It has been also applied to corporate credit ratings. In this study we utilized DEA for sorting venture companies by efficiency based ratings. The Support Vector Machine(SVM), on the other hand, is a popular technique for solving data classification problems. In this paper, we employed SVM to classify the efficiency ratings in IT venture companies according to the results of DEA. The SVM method was first developed by Vapnik (1995). As one of many machine learning techniques, SVM is based on a statistical theory. Thus far, the method has shown good performances especially in generalizing capacity in classification tasks, resulting in numerous applications in many areas of business, SVM is basically the algorithm that finds the maximum margin hyperplane, which is the maximum separation between classes. According to this method, support vectors are the closest to the maximum margin hyperplane. If it is impossible to classify, we can use the kernel function. In the case of nonlinear class boundaries, we can transform the inputs into a high-dimensional feature space, This is the original input space and is mapped into a high-dimensional dot-product space. Many studies applied SVM to the prediction of bankruptcy, the forecast a financial time series, and the problem of estimating credit rating, In this study we employed SVM for developing data mining-based efficiency prediction model. We used the Gaussian radial function as a kernel function of SVM. In multi-class SVM, we adopted one-against-one approach between binary classification method and two all-together methods, proposed by Weston and Watkins(1999) and Crammer and Singer(2000), respectively. In this research, we used corporate information of 154 companies listed on KOSDAQ market in Korea exchange. We obtained companies' financial information of 2005 from the KIS(Korea Information Service, Inc.). Using this data, we made multi-class rating with DEA efficiency and built multi-class prediction model based data mining. Among three manners of multi-classification, the hit ratio of the Weston and Watkins method is the best in the test data set. In multi classification problems as efficiency ratings of venture business, it is very useful for investors to know the class with errors, one class difference, when it is difficult to find out the accurate class in the actual market. So we presented accuracy results within 1-class errors, and the Weston and Watkins method showed 85.7% accuracy in our test samples. We conclude that the DEA based multi-class approach in venture business generates more information than the binary classification problem, notwithstanding its efficiency level. We believe this model can help investors in decision making as it provides a reliably tool to evaluate venture companies in the financial domain. For the future research, we perceive the need to enhance such areas as the variable selection process, the parameter selection of kernel function, the generalization, and the sample size of multi-class.

• Journal of Life Science
• /
• v.17 no.12
• /
• pp.1641-1648
• /
• 2007

Yippee-like proteins, which have been identified as the homolog of Drosophila yippee protein containing a zinc-finger domain, are known to be highly conserved among eukaryotes. However, their functional roles are still poorly understood. Recently we initiated ordered differential display (ODD)-polymerase chain reaction (PCR) to isolate genes of which expressions are altered following activation of human T cells. On the ODD-PCR image, one PCR-product detected in unstimulated T cells was not detectable at the time when the activated T cells traversed near $G_1/S$ boundary following activation by immobilized anti-CD3. Cloning and nucleotide sequence analysis revealed that the PCR-product was yippee-like 5 (YPEL5) gene, which was known as a human homolog of the Drosophila yippee gene. Northern blot analysis confirmed the amount of ${\sim}2.2$ kb YPEL5 mRNA expression detectable in unstimulated T cells was sustained until 1.5 hr after activation and then rapidly declined to undetectable level by 5 hr. Ectopic expression of YPEL5 gene in human cervix epitheloid carcinoma HeLa cells caused a significant reduction in cell proliferation to the level of 47% of the control. Expression of GFP-YPEL5 fusion protein in HeLa cells showed its nuclear localization. These results demonstrated that the expression level of human YPEL5 mRNA was negatively regulated in the early stage of T cell activation, and suggested that YPEL5 might exert an inhibitory effect on the cell proliferation as a nuclear protein.

• Journal of Intelligence and Information Systems
• /
• v.19 no.1
• /
• pp.125-139
• /
• 2013

Big data refers to the data that cannot be processes with conventional contemporary data technologies. As smart devices and social network services produces vast amount of data, big data attracts much attention from researchers. There are strong demands form governments and industries for bib data as it can create new values by drawing business insights from data. Since various new technologies to process big data introduced, academic communities also show much interest to the big data domain. A notable advance related to the big data technology has been in various fields. Big data technology makes it possible to access, collect, and save individual's personal data. These technologies enable the analysis of huge amounts of data with lower cost and less time, which is impossible to achieve with traditional methods. It even detects personal information that people do not want to open. Therefore, people using information technology such as the Internet or online services have some level of privacy concerns, and such feelings can hinder continued use of information systems. For example, SNS offers various benefits, but users are sometimes highly exposed to privacy intrusions because they write too much personal information on it. Even though users post their personal information on the Internet by themselves, the data sometimes is not under control of the users. Once the private data is posed on the Internet, it can be transferred to anywhere by a few clicks, and can be abused to create fake identity. In this way, privacy intrusion happens. This study aims to investigate how perceived personal information overload in SNS affects user's risk perception and information privacy concerns. Also, it examines the relationship between the concerns and user resistance behavior. A survey approach and structural equation modeling method are employed for data collection and analysis. This study contributes meaningful insights for academic researchers and policy makers who are planning to develop guidelines for privacy protection. The study shows that information overload on the social network services can bring the significant increase of users' perceived level of privacy risks. In turn, the perceived privacy risks leads to the increased level of privacy concerns. IF privacy concerns increase, it can affect users to from a negative or resistant attitude toward system use. The resistance attitude may lead users to discontinue the use of social network services. Furthermore, information overload is mediated by perceived risks to affect privacy concerns rather than has direct influence on perceived risk. It implies that resistance to the system use can be diminished by reducing perceived risks of users. Given that users' resistant behavior become salient when they have high privacy concerns, the measures to alleviate users' privacy concerns should be conceived. This study makes academic contribution of integrating traditional information overload theory and user resistance theory to investigate perceived privacy concerns in current IS contexts. There is little big data research which examined the technology with empirical and behavioral approach, as the research topic has just emerged. It also makes practical contributions. Information overload connects to the increased level of perceived privacy risks, and discontinued use of the information system. To keep users from departing the system, organizations should develop a system in which private data is controlled and managed with ease. This study suggests that actions to lower the level of perceived risks and privacy concerns should be taken for information systems continuance.

• Korean Journal of Soil Science and Fertilizer
• /
• v.43 no.6
• /
• pp.804-811
• /
• 2010

The lower portion of sloping paddy fields normally contains excessive moisture and the higher water table caused by the inflow of ground water from the upper part of the field resulting in non-uniform water content distribution. Four drainage methods namely Open Ditch, Vinyl Barrier, Pipe Drainage and Tube Bundle for multiple land use were installed within 1-m position from the lower edge of the upper embankment of sloping alluvial paddy fields. Knowledge of the spatial variability of soil water properties is of primary importance for management of agricultural lands. This study was conducted to evaluate the effect of drainage in the soil on spatial variability of soil water content using the geostatistical analysis. The soil water content was collected by a TDR (Time Domain Reflectometry) sensor after the installation of subsurface drainage on regular square grid of 80 m at 20 m paddy field located at Oesan-ri, Buk-myeon, Changwon-si in alluvial slopping paddy fields ($35^{\circ}22^{\prime}$ N, $128^{\circ}35^{\prime}$). In order to obtain the most accurate field information, the sampling grid was divided 3 m by 3 m unit mesh by four drainage types. The results showed that spatial variance of soil water content by subsurface drainage was reduced, though yield of soybean showed the same trends. Value of "sill" of soil water content with semivariogram was 9.7 in Pipe Drainage, 86.2 in Open Ditch, and 66.8 in Vinyl Barrier and 15.7 in Tube Bundle.

• Asia-Pacific Journal of Business Venturing and Entrepreneurship
• /
• v.18 no.1
• /
• pp.67-83
• /
• 2023

The social venture start-up phenomenon is found from the perspectives of social enterprise and for-profit enterprise. This study aims to fundamentally explore the start-up phenomenon of social ventures from these two perspectives. Considering the lack of prior research that researched both social and commercial perspectives at the same time, this paper analyzed using grounded theory approach of Strauss & Corbin(1998), an inductive research method that analyzes based on prior research and interview data. In order to collect data for this study, eight corporate representatives currently operating social ventures were interviewed and data and phenomena were analyzed. This progressed to a theoretical saturation where no additional information was derived. The analysis results of this study using the grounded theory approach are as follows. As a result of open coding and axial coding, 147 concepts and 70 subcategories were derived, and 18 categories were derived through the final abstraction process. In the selective coding, 'expansion of social venture entry in the social domain' and 'expansion of social function of for-profit companies' were selected as key categories, and a story line was formed around this. In this study, we saw that it is necessary to conduct academic research and analysis on the competitive factors required for companies that pursue the values of two conflicting relationships, such as social ventures, to survive with competitiveness. In practice, concepts such as collaboration with for-profit companies, value combination, entrepreneurship competency and performance improvement, social value execution competency reinforcement, communication strategy, for-profit enterprise value investment, and entrepreneur management competency were derived. This study explains the social venture phenomenon for social enterprises, commercial enterprises, and entrepreneurs who want to enter the social venture field. It is expected to provide the implications necessary for successful social venture startups.