• Journal of Intelligence and Information Systems
• /
• v.22 no.2
• /
• pp.127-142
• /
• 2016

Deep learning model is a kind of neural networks that allows multiple hidden layers. There are various deep learning architectures such as convolutional neural networks, deep belief networks and recurrent neural networks. Those have been applied to fields like computer vision, automatic speech recognition, natural language processing, audio recognition and bioinformatics where they have been shown to produce state-of-the-art results on various tasks. Among those architectures, convolutional neural networks and recurrent neural networks are classified as the supervised learning model. And in recent years, those supervised learning models have gained more popularity than unsupervised learning models such as deep belief networks, because supervised learning models have shown fashionable applications in such fields mentioned above. Deep learning models can be trained with backpropagation algorithm. Backpropagation is an abbreviation for "backward propagation of errors" and a common method of training artificial neural networks used in conjunction with an optimization method such as gradient descent. The method calculates the gradient of an error function with respect to all the weights in the network. The gradient is fed to the optimization method which in turn uses it to update the weights, in an attempt to minimize the error function. Convolutional neural networks use a special architecture which is particularly well-adapted to classify images. Using this architecture makes convolutional networks fast to train. This, in turn, helps us train deep, muti-layer networks, which are very good at classifying images. These days, deep convolutional networks are used in most neural networks for image recognition. Convolutional neural networks use three basic ideas: local receptive fields, shared weights, and pooling. By local receptive fields, we mean that each neuron in the first(or any) hidden layer will be connected to a small region of the input(or previous layer's) neurons. Shared weights mean that we're going to use the same weights and bias for each of the local receptive field. This means that all the neurons in the hidden layer detect exactly the same feature, just at different locations in the input image. In addition to the convolutional layers just described, convolutional neural networks also contain pooling layers. Pooling layers are usually used immediately after convolutional layers. What the pooling layers do is to simplify the information in the output from the convolutional layer. Recent convolutional network architectures have 10 to 20 hidden layers and billions of connections between units. Training deep learning networks has taken weeks several years ago, but thanks to progress in GPU and algorithm enhancement, training time has reduced to several hours. Neural networks with time-varying behavior are known as recurrent neural networks or RNNs. A recurrent neural network is a class of artificial neural network where connections between units form a directed cycle. This creates an internal state of the network which allows it to exhibit dynamic temporal behavior. Unlike feedforward neural networks, RNNs can use their internal memory to process arbitrary sequences of inputs. Early RNN models turned out to be very difficult to train, harder even than deep feedforward networks. The reason is the unstable gradient problem such as vanishing gradient and exploding gradient. The gradient can get smaller and smaller as it is propagated back through layers. This makes learning in early layers extremely slow. The problem actually gets worse in RNNs, since gradients aren't just propagated backward through layers, they're propagated backward through time. If the network runs for a long time, that can make the gradient extremely unstable and hard to learn from. It has been possible to incorporate an idea known as long short-term memory units (LSTMs) into RNNs. LSTMs make it much easier to get good results when training RNNs, and many recent papers make use of LSTMs or related ideas.

• Journal of Aquaculture
• /
• v.21 no.4
• /
• pp.317-324
• /
• 2008

This study estimated the genetic parameters and breeding values for the growth-related traits of the 1st generation produced in 2005. The heritability of body weight, total length, body height, body shape and condition factor of 180 days old olive flounders Paralichthys olivaceus, the 1st generation of selection, was estimated as 0.564, 0.590, 0.588, 0.306 and 0.332, respectively. And reproductivity of genetic evaluation for crossing superior flounders and inferior ones was tested using the subsequent generation produced in 2006 based on the breeding values of 1st generation. In the least-squares means of body weight and total length for each group of crossing, the values of crossing group between superior flounders ($S{\times}S$) showed $145.6{\pm}1.8\;g$ and $22.4{\pm}0.1\;cm$, respectively. The values of crossing group between superior and inferior flounders ($S{\times}I$) showed $133.2{\pm}2.5\;g$ and $22.1{\pm}0.1\;cm$, respectively. The values of crossing group between inferior flounders ($I{\times}I$) showed $114.0{\pm}2.1\;g$ and $21.08{\pm}0.12\;cm$, respectively. In the results, flounders are determined as appropriate selective breeding fish with the high heritability of flounders in early ages at 180 days old, and the reproductivity of genetic evaluation was also high.

• The Journal of Korean Society for Radiation Therapy
• /
• v.26 no.2
• /
• pp.313-320
• /
• 2014

Purpose : The purpose of the study is to evaluate the efficiency of Coaxial MLC VMAT plan (Using $273^{\circ}$ and $350^{\circ}$ collimator angle) That the leaf motion direction aligned with axis of OAR (Organ at risk, It means spinal cord or cauda equine in this study.) compare to Universal MLC VMAT plan (using $30^{\circ}$ and $330^{\circ}$ collimator angle) for spine SBRT. Materials and Methods : The 10 cases of spine SBRT that treated with VMAT planned by Coaxial MLC and Varian TBX were enrolled. Those cases were planned by Eclipse (Ver. 10.0.42, Varian, USA), PRO3 (Progressive Resolution Optimizer 10.0.28) and AAA (Anisotropic Analytic Algorithm Ver. 10.0.28) with coplanar $360^{\circ}$ arcs and 10MV FFF (Flattening filter free). Each arc has $273^{\circ}$ and $350^{\circ}$ collimator angle, respectively. The Universal MLC VMAT plans are based on existing treatment plans. Those plans have the same parameters of existing treatment plans but collimator angle. To minimize the dose difference that shows up randomly on optimizing, all plans were optimized and calculated twice respectively. The calculation grid is 0.2 cm and all plans were normalized to the target V100%=90%. The indexes of evaluation are V10Gy, D0.03cc, Dmean of OAR (Organ at risk, It means spinal cord or cauda equine in this study.), H.I (Homogeneity index) of the target and total MU. All Coaxial VMAT plans were verified by gamma test with Mapcheck2 (Sun Nuclear Co., USA), Mapphan (Sun Nuclear Co., USA) and SNC patient (Sun Nuclear Co., USA Ver 6.1.2.18513). Results : The difference between the coaxial and the universal VMAT plans are follow. The coaxial VMAT plan is better in the V10Gy of OAR, Up to 4.1%, at least 0.4%, the average difference was 1.9% and In the D0.03cc of OAR, Up to 83.6 cGy, at least 2.2 cGy, the average difference was 33.3 cGy. In Dmean, Up to 34.8 cGy, at least -13.0 cGy, the average difference was 9.6 cGy that say the coaxial VMAT plans are better except few cases. H.I difference Up to 0.04, at least 0.01, the average difference was 0.02 and the difference of average total MU is 74.1 MU. The coaxial MLC VMAT plan is average 74.1 MU lesser then another. All IMRT verification gamma test results for the coaxial MLC VMAT plan passed over 90.0% at 1mm / 2%. Conclusion : Coaxial MLC VMAT treatment plan appeared to be favorable in most cases than the Universal MLC VMAT treatment planning. It is efficient in lowering the dose of the OAR V10Gy especially. As a result, the Coaxial MLC VMAT plan could be better than the Universal MLC VMAT plan in same MU.

• Korean Journal of Remote Sensing
• /
• v.36 no.2_2
• /
• pp.379-399
• /
• 2020

The detection of illegal ship is one of the key factors in building a marine surveillance system. Effective marine surveillance requires the means for continuous monitoring over a wide area. In this study, the possibility of ship detection monitoring based on satellite SAR, HF radar, UAV and AIS integration was investigated. Considering the characteristics of time and spatial resolution for each platform, the ship monitoring scenario consisted of a regular surveillance system using HFR data and AIS data, and an event monitoring system using satellites and UAVs. The regular surveillance system still has limitations in detecting a small ship and accuracy due to the low spatial resolution of HF radar data. However, the event monitoring system using satellite SAR data effectively detects illegal ships using AIS data, and the ship speed and heading direction estimated from SAR images or ship tracking information using HF radar data can be used as the main information for the transition to UAV monitoring. For the validation of monitoring scenario, a comprehensive field experiment was conducted from June 25 to June 26, 2019, at the west side of Hongwon Port in Seocheon. KOMPSAT-5 SAR images, UAV data, HF radar data and AIS data were successfully collected and analyzed by applying each developed algorithm. The developed system will be the basis for the regular and event ship monitoring scenarios as well as the visualization of data and analysis results collected from multiple platforms.

• Journal of Intelligence and Information Systems
• /
• v.21 no.1
• /
• pp.29-45
• /
• 2015

Response modeling is a well-known research issue for those who have tried to get more superior performance in the capability of predicting the customers' response for the marketing promotion. The response model for customers would reduce the marketing cost by identifying prospective customers from very large customer database and predicting the purchasing intention of the selected customers while the promotion which is derived from an undifferentiated marketing strategy results in unnecessary cost. In addition, the big data environment has accelerated developing the response model with data mining techniques such as CBR, neural networks and support vector machines. And CBR is one of the most major tools in business because it is known as simple and robust to apply to the response model. However, CBR is an attractive data mining technique for data mining applications in business even though it hasn't shown high performance compared to other machine learning techniques. Thus many studies have tried to improve CBR and utilized in business data mining with the enhanced algorithms or the support of other techniques such as genetic algorithm, decision tree and AHP (Analytic Process Hierarchy). Ahn and Kim(2008) utilized logit, neural networks, CBR to predict that which customers would purchase the items promoted by marketing department and tried to optimized the number of k for k-nearest neighbor with genetic algorithm for the purpose of improving the performance of the integrated model. Hong and Park(2009) noted that the integrated approach with CBR for logit, neural networks, and Support Vector Machine (SVM) showed more improved prediction ability for response of customers to marketing promotion than each data mining models such as logit, neural networks, and SVM. This paper presented an approach to predict customers' response of marketing promotion with Case Based Reasoning. The proposed model was developed by applying different weights to each feature. We deployed logit model with a database including the promotion and the purchasing data of bath soap. After that, the coefficients were used to give different weights of CBR. We analyzed the performance of proposed weighted CBR based model compared to neural networks and pure CBR based model empirically and found that the proposed weighted CBR based model showed more superior performance than pure CBR model. Imbalanced data is a common problem to build data mining model to classify a class with real data such as bankruptcy prediction, intrusion detection, fraud detection, churn management, and response modeling. Imbalanced data means that the number of instance in one class is remarkably small or large compared to the number of instance in other classes. The classification model such as response modeling has a lot of trouble to recognize the pattern from data through learning because the model tends to ignore a small number of classes while classifying a large number of classes correctly. To resolve the problem caused from imbalanced data distribution, sampling method is one of the most representative approach. The sampling method could be categorized to under sampling and over sampling. However, CBR is not sensitive to data distribution because it doesn't learn from data unlike machine learning algorithm. In this study, we investigated the robustness of our proposed model while changing the ratio of response customers and nonresponse customers to the promotion program because the response customers for the suggested promotion is always a small part of nonresponse customers in the real world. We simulated the proposed model 100 times to validate the robustness with different ratio of response customers to response customers under the imbalanced data distribution. Finally, we found that our proposed CBR based model showed superior performance than compared models under the imbalanced data sets. Our study is expected to improve the performance of response model for the promotion program with CBR under imbalanced data distribution in the real world.

• Journal of Life Science
• /
• v.22 no.3
• /
• pp.366-372
• /
• 2012

This study was conducted to estimate the extent of linkage disequilibrium (LD) and effective population size using whole genomic single nucleotide polymorphisms (SNP) genotyped by DNA chip in Hanwoo. Using the blood samples of 35 young bulls born from 2005 to 2008 and their progenies (N=253) in a Hanwoo nucleus population collected from Hanwoo Improvement Center, 51,582 SNPs were genotyped using Bovine SNP50 chips. A total of 40,851 SNPs were used in this study after elimination of SNPs with a missing genotyping rate of over 10 percent and monomorphic SNPs (10,730 SNPs). The total autosomal genome length, measured as the sum of the longest syntenic pairs of SNPs by chromosome, was 2,541.6 Mb (Mega base pairs). The average distances of all adjacent pairs by each BTA ranged from 0.55 to 0.74 cM. Decay of LD showed an exponential trend with physical distance. The means of LD ($r^2$) among syntenic SNP pairs were 0.136 at a range of 0-0.1 Mb in physical distance and 0.06 at a range of 0.1-0.2 Mb. When these results were used for Luo's formula, about 2,000 phenotypic records were found to be required to achieve power > 0.9 to detect 5% QTL in the population of Hanwoo. As a result of estimating effective population size by generation in Hanwoo, the estimated effective population size for the current status was 84 heads and the estimate of effective population size for 50 generations of ancestors was 1,150 heads. The average decreasing rates of effective population size by generation were 9.0% at about five generations and 17.3% at the current generation. The main cause of the rapid decrease in effective population size was considered to be the intensive use of a few prominent sires since the application of artificial insemination technology in Korea. To increase and/or sustain the effective population size, the selection of various proven bulls and mating systems that consider genetic diversity are needed.

• Journal of Intelligence and Information Systems
• /
• v.19 no.4
• /
• pp.147-157
• /
• 2013

With the advent of the digital environment that can be accessed anytime, anywhere with the introduction of high-speed network, the free distribution and use of digital content were made possible. Ironically this environment is raising a variety of copyright infringement, and product images used in the online shopping mall are pirated frequently. There are many controversial issues whether shopping mall images are creative works or not. According to Supreme Court's decision in 2001, to ad pictures taken with ham products is simply a clone of the appearance of objects to deliver nothing but the decision was not only creative expression. But for the photographer's losses recognized in the advertising photo shoot takes the typical cost was estimated damages. According to Seoul District Court precedents in 2003, if there are the photographer's personality and creativity in the selection of the subject, the composition of the set, the direction and amount of light control, set the angle of the camera, shutter speed, shutter chance, other shooting methods for capturing, developing and printing process, the works should be protected by copyright law by the Court's sentence. In order to receive copyright protection of the shopping mall images by the law, it is simply not to convey the status of the product, the photographer's personality and creativity can be recognized that it requires effort. Accordingly, the cost of making the mall image increases, and the necessity for copyright protection becomes higher. The product images of the online shopping mall have a very unique configuration unlike the general pictures such as portraits and landscape photos and, therefore, the general image watermarking technique can not satisfy the requirements of the image watermarking. Because background of product images commonly used in shopping malls is white or black, or gray scale (gradient) color, it is difficult to utilize the space to embed a watermark and the area is very sensitive even a slight change. In this paper, the characteristics of images used in shopping malls are analyzed and a watermarking technology which is suitable to the shopping mall images is proposed. The proposed image watermarking technology divide a product image into smaller blocks, and the corresponding blocks are transformed by DCT (Discrete Cosine Transform), and then the watermark information was inserted into images using quantization of DCT coefficients. Because uniform treatment of the DCT coefficients for quantization cause visual blocking artifacts, the proposed algorithm used weighted mask which quantizes finely the coefficients located block boundaries and coarsely the coefficients located center area of the block. This mask improves subjective visual quality as well as the objective quality of the images. In addition, in order to improve the safety of the algorithm, the blocks which is embedded the watermark are randomly selected and the turbo code is used to reduce the BER when extracting the watermark. The PSNR(Peak Signal to Noise Ratio) of the shopping mall image watermarked by the proposed algorithm is 40.7~48.5[dB] and BER(Bit Error Rate) after JPEG with QF = 70 is 0. This means the watermarked image is high quality and the algorithm is robust to JPEG compression that is used generally at the online shopping malls. Also, for 40% change in size and 40 degrees of rotation, the BER is 0. In general, the shopping malls are used compressed images with QF which is higher than 90. Because the pirated image is used to replicate from original image, the proposed algorithm can identify the copyright infringement in the most cases. As shown the experimental results, the proposed algorithm is suitable to the shopping mall images with simple background. However, the future study should be carried out to enhance the robustness of the proposed algorithm because the robustness loss is occurred after mask process.

• Journal of Internet Computing and Services
• /
• v.16 no.2
• /
• pp.77-83
• /
• 2015

Frequent pattern mining, which is one of the major areas actively studied in data mining, is a method for extracting useful pattern information hidden from large data sets or databases. Moreover, frequent pattern mining approaches have been actively employed in a variety of application fields because the results obtained from them can allow us to analyze various, important characteristics within databases more easily and automatically. However, traditional frequent pattern mining methods, which simply extract all of the possible frequent patterns such that each of their support values is not smaller than a user-given minimum support threshold, have the following problems. First, traditional approaches have to generate a numerous number of patterns according to the features of a given database and the degree of threshold settings, and the number can also increase in geometrical progression. In addition, such works also cause waste of runtime and memory resources. Furthermore, the pattern results excessively generated from the methods also lead to troubles of pattern analysis for the mining results. In order to solve such issues of previous traditional frequent pattern mining approaches, the concept of representative pattern mining and its various related works have been proposed. In contrast to the traditional ones that find all the possible frequent patterns from databases, representative pattern mining approaches selectively extract a smaller number of patterns that represent general frequent patterns. In this paper, we describe details and characteristics of pattern condensing techniques that consider the maximality or closure property of generated frequent patterns, and conduct comparison and analysis for the techniques. Given a frequent pattern, satisfying the maximality for the pattern signifies that all of the possible super sets of the pattern must have smaller support values than a user-specific minimum support threshold; meanwhile, satisfying the closure property for the pattern means that there is no superset of which the support is equal to that of the pattern with respect to all the possible super sets. By mining maximal frequent patterns or closed frequent ones, we can achieve effective pattern compression and also perform mining operations with much smaller time and space resources. In addition, compressed patterns can be converted into the original frequent pattern forms again if necessary; especially, the closed frequent pattern notation has the ability to convert representative patterns into the original ones again without any information loss. That is, we can obtain a complete set of original frequent patterns from closed frequent ones. Although the maximal frequent pattern notation does not guarantee a complete recovery rate in the process of pattern conversion, it has an advantage that can extract a smaller number of representative patterns more quickly compared to the closed frequent pattern notation. In this paper, we show the performance results and characteristics of the aforementioned techniques in terms of pattern generation, runtime, and memory usage by conducting performance evaluation with respect to various real data sets collected from the real world. For more exact comparison, we also employ the algorithms implementing these techniques on the same platform and Implementation level.

• Journal of Intelligence and Information Systems
• /
• v.19 no.3
• /
• pp.57-71
• /
• 2013

Over a billion people in the world generate new news minute by minute. People forecasts some news but most news are from unexpected events such as natural disasters, accidents, crimes. People spend much time to watch a huge amount of news delivered from many media because they want to understand what is happening now, to predict what might happen in the near future, and to share and discuss on the news. People make better daily decisions through watching and obtaining useful information from news they saw. However, it is difficult that people choose news suitable to them and obtain useful information from the news because there are so many news media such as portal sites, broadcasters, and most news articles consist of gossipy news and breaking news. User interest changes over time and many people have no interest in outdated news. From this fact, applying users' recent interest to personalized news service is also required in news service. It means that personalized news service should dynamically manage user profiles. In this paper, a content-based news recommendation system is proposed to provide the personalized news service. For a personalized service, user's personal information is requisitely required. Social network service is used to extract user information for personalization service. The proposed system constructs dynamic user profile based on recent user information of Facebook, which is one of social network services. User information contains personal information, recent articles, and Facebook Page information. Facebook Pages are used for businesses, organizations and brands to share their contents and connect with people. Facebook users can add Facebook Page to specify their interest in the Page. The proposed system uses this Page information to create user profile, and to match user preferences to news topics. However, some Pages are not directly matched to news topic because Page deals with individual objects and do not provide topic information suitable to news. Freebase, which is a large collaborative database of well-known people, places, things, is used to match Page to news topic by using hierarchy information of its objects. By using recent Page information and articles of Facebook users, the proposed systems can own dynamic user profile. The generated user profile is used to measure user preferences on news. To generate news profile, news category predefined by news media is used and keywords of news articles are extracted after analysis of news contents including title, category, and scripts. TF-IDF technique, which reflects how important a word is to a document in a corpus, is used to identify keywords of each news article. For user profile and news profile, same format is used to efficiently measure similarity between user preferences and news. The proposed system calculates all similarity values between user profiles and news profiles. Existing methods of similarity calculation in vector space model do not cover synonym, hypernym and hyponym because they only handle given words in vector space model. The proposed system applies WordNet to similarity calculation to overcome the limitation. Top-N news articles, which have high similarity value for a target user, are recommended to the user. To evaluate the proposed news recommendation system, user profiles are generated using Facebook account with participants consent, and we implement a Web crawler to extract news information from PBS, which is non-profit public broadcasting television network in the United States, and construct news profiles. We compare the performance of the proposed method with that of benchmark algorithms. One is a traditional method based on TF-IDF. Another is 6Sub-Vectors method that divides the points to get keywords into six parts. Experimental results demonstrate that the proposed system provide useful news to users by applying user's social network information and WordNet functions, in terms of prediction error of recommended news.

• Journal of Intelligence and Information Systems
• /
• v.27 no.1
• /
• pp.23-46
• /
• 2021

Collaborative filtering, which is often used in personalization recommendations, is recognized as a very useful technique to find similar customers and recommend products to them based on their purchase history. However, the traditional collaborative filtering technique has raised the question of having difficulty calculating the similarity for new customers or products due to the method of calculating similaritiesbased on direct connections and common features among customers. For this reason, a hybrid technique was designed to use content-based filtering techniques together. On the one hand, efforts have been made to solve these problems by applying the structural characteristics of social networks. This applies a method of indirectly calculating similarities through their similar customers placed between them. This means creating a customer's network based on purchasing data and calculating the similarity between the two based on the features of the network that indirectly connects the two customers within this network. Such similarity can be used as a measure to predict whether the target customer accepts recommendations. The centrality metrics of networks can be utilized for the calculation of these similarities. Different centrality metrics have important implications in that they may have different effects on recommended performance. In this study, furthermore, the effect of these centrality metrics on the performance of recommendation may vary depending on recommender algorithms. In addition, recommendation techniques using network analysis can be expected to contribute to increasing recommendation performance even if they apply not only to new customers or products but also to entire customers or products. By considering a customer's purchase of an item as a link generated between the customer and the item on the network, the prediction of user acceptance of recommendation is solved as a prediction of whether a new link will be created between them. As the classification models fit the purpose of solving the binary problem of whether the link is engaged or not, decision tree, k-nearest neighbors (KNN), logistic regression, artificial neural network, and support vector machine (SVM) are selected in the research. The data for performance evaluation used order data collected from an online shopping mall over four years and two months. Among them, the previous three years and eight months constitute social networks composed of and the experiment was conducted by organizing the data collected into the social network. The next four months' records were used to train and evaluate recommender models. Experiments with the centrality metrics applied to each model show that the recommendation acceptance rates of the centrality metrics are different for each algorithm at a meaningful level. In this work, we analyzed only four commonly used centrality metrics: degree centrality, betweenness centrality, closeness centrality, and eigenvector centrality. Eigenvector centrality records the lowest performance in all models except support vector machines. Closeness centrality and betweenness centrality show similar performance across all models. Degree centrality ranking moderate across overall models while betweenness centrality always ranking higher than degree centrality. Finally, closeness centrality is characterized by distinct differences in performance according to the model. It ranks first in logistic regression, artificial neural network, and decision tree withnumerically high performance. However, it only records very low rankings in support vector machine and K-neighborhood with low-performance levels. As the experiment results reveal, in a classification model, network centrality metrics over a subnetwork that connects the two nodes can effectively predict the connectivity between two nodes in a social network. Furthermore, each metric has a different performance depending on the classification model type. This result implies that choosing appropriate metrics for each algorithm can lead to achieving higher recommendation performance. In general, betweenness centrality can guarantee a high level of performance in any model. It would be possible to consider the introduction of proximity centrality to obtain higher performance for certain models.