• Investigative Magnetic Resonance Imaging
• /
• v.25 no.1
• /
• pp.10-22
• /
• 2021

Purpose: To overcome the difficulty in building a large data set with a high-quality in medical imaging, a concept of 'blended-transfer learning' (BTL) using a combination of both source data and target data is proposed for the target task. Materials and Methods: Source and target tasks were defined as training of the source and target networks to reconstruct cardiac CINE images from undersampled data, respectively. In transfer learning (TL), the entire neural network (NN) or some parts of the NN after conducting a source task using an open data set was adopted in the target network as the initial network to improve the learning speed and the performance of the target task. Using BTL, an NN effectively learned the target data while preserving knowledge from the source data to the maximum extent possible. The ratio of the source data to the target data was reduced stepwise from 1 in the initial stage to 0 in the final stage. Results: NN that performed BTL showed an improved performance compared to those that performed TL or standalone learning (SL). Generalization of NN was also better achieved. The learning curve was evaluated using normalized mean square error (NMSE) of reconstructed images for both target data and source data. BTL reduced the learning time by 1.25 to 100 times and provided better image quality. Its NMSE was 3% to 8% lower than with SL. Conclusion: The NN that performed the proposed BTL showed the best performance in terms of learning speed and learning curve. It also showed the highest reconstructed-image quality with the lowest NMSE for the test data set. Thus, BTL is an effective way of learning for NNs in the medical-imaging domain where both quality and quantity of data are always limited.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.8 no.5
• /
• pp.1108-1113
• /
• 2004

‘Set’ is a tool that is used frequently in designing computer programs. Because of the reason, ‘set constraints languages’ have been developed recently. In this research, we introduce ‘(Ab)(Cl) set unification’ problem and implement it using the ‘set equation rewriting in Prolog’. In this research we shows that the set unification, that is considered to be difficult to be implemented in procedural languages, ran be implemented easily using the non-deterministic control structure and the list data structure in logic language like Prolog. Our research uses the Ciao Prolog with GNU GPL, this is compared with other existing implementations which used expensive commercial Prolog, so anyone can use the result freely. Currently the result is being used for implementing a set constraint language.

• Proceedings of the Korean Information Science Society Conference
• /
• 2004.10b
• /
• pp.703-705
• /
• 2004

Conventional support vector machines (SVMs) find optimal hyperplanes that have maximal margins by treating all data equivalently. In the real world, however, the data within a data set may differ in degree of uncertainty or importance due to noise, inaccuracies or missing values in the data. Hence, if all data are treated as equivalent, without considering such differences, the optimal hyperplanes identified are likely to be less optimal. In this paper, to more accurately identify the optimal hyperplane in a given uncertain data set, we propose a membership-induced distance from a hyperplane using membership values, and formulate three kinds of membership-induced SVMs.

• Journal of the Korean Data and Information Science Society
• /
• v.14 no.2
• /
• pp.345-354
• /
• 2003

An incremental kernel principal component analysis(IKPCA) is proposed for the nonlinear feature extraction from the data. The problem of batch kernel principal component analysis(KPCA) is that the computation becomes prohibitive when the data set is large. Another problem is that, in order to update the eigenvectors with another data, the whole eigenvectors should be recomputed. IKPCA overcomes this problem by incrementally updating the eigenspace model. IKPCA is more efficient in memory requirement than a batch KPCA and can be easily improved by re-learning the data. In our experiments we show that IKPCA is comparable in performance to a batch KPCA for the classification problem on nonlinear data set.

• Journal of Korea Multimedia Society
• /
• v.12 no.10
• /
• pp.1457-1465
• /
• 2009

In MMORPG development process, spreadsheet based data management has advantages in the handling of functions and analysis of large data set. But it has limitations in inserting, deleting, searching and relational management of that. In this paper, we proposed the spreadsheet based tool model for large data set for MMORPG development. Our system can reduce the risk of data management failure in MMORPG development process and improve the efficiency of data handling in the large-scale team.

• Food Science and Biotechnology
• /
• v.18 no.1
• /
• pp.137-142
• /
• 2009

One of the most important aspects of conducting this microbial risk assessment (MRA) is determining the model in microbial behaviors in food systems. However, to fully these modeling, large expenditures or newly laboratory experiments will be spent to do it. To overcome these problems, it has to be considered to develop the new strategies that can be used data in the published literatures. This study is to show whether or not the data set from the published experimental data has more value for modeling for MRA. To illustrate this suggestion, as example of data set, 4 published Salmonella survival in Cheddar cheese reports were used. Finally, using the GInaFiT tool, survival was modeled by nonlinear polynomial regression model describing the effect of temperature on Weibull model parameters. This model used data in the literatures is useful in describing behavior of Salmonella during different time and temperature conditions of cheese ripening.

• Journal of Intelligence and Information Systems
• /
• v.5 no.1
• /
• pp.95-101
• /
• 1999

In this paper, a comparison between different methods to combine predictions from neural networks will be given. These methods are bagging, bumping, and balancing. Those are based on the analysis of the ensemble generalization error into an ambiguity term and a term incorporating generalization performances of individual networks. Neural Networks and AI machine learning models are prone to overfitting. A strategy to prevent a neural network from overfitting, is to stop training in early stage of the learning process. The complete data set is spilt up into a training set and a validation set. Training is stopped when the error on the validation set starts increasing. The stability of the networks is highly dependent on the division in training and validation set, and also on the random initial weights and the chosen minimization procedure. This causes early stopped networks to be rather unstable: a small change in the data or different initial conditions can produce large changes in the prediction. Therefore, it is advisable to apply the same procedure several times starting from different initial weights. This technique is often referred to as training ensembles of neural networks. In this paper, we presented a comparison of three statistical methods to prevent overfitting of neural network.

• Proceedings of the KIEE Conference
• /
• 2005.05a
• /
• pp.3-5
• /
• 2005

In this paper. we introduce a new structure of fuzzy-neural networks Fuzzy Set-based Polynomial Neural Networks (FSPNN). The two underlying design mechanisms of such networks involve genetic optimization and information granulation. The resulting constructs are Fuzzy Polynomial Neural Networks (FPNN) with fuzzy set-based polynomial neurons (FSPNs) regarded as their generic processing elements. First, we introduce a comprehensive design methodology (viz. a genetic optimization using Genetic Algorithms) to determine the optimal structure of the FSPNNs. This methodology hinges on the extended Group Method of Data Handling (GMDH) and fuzzy set-based rules. It concerns FSPNN-related parameters such as the number of input variables, the order of the polynomial, the number of membership functions, and a collection of a specific subset of input variables realized through the mechanism of genetic optimization. Second, the fuzzy rules used in the networks exploit the notion of information granules defined over systems variables and formed through the process of information granulation. This granulation is realized with the aid of the hard C-Means clustering (HCM). The performance of the network is quantified through experimentation in which we use a number of modeling benchmarks already experimented with in the realm of fuzzy or neurofuzzy modeling.

• Journal of Information Technology Applications and Management
• /
• v.12 no.3
• /
• pp.27-39
• /
• 2005

The Semantic Web has increased the interest in ontologies recently Ontology is an essential component of the semantic web and continues to change and evolve. We consider versions management schemes in ontology. We study a set of changes based on domain changes, changes in conceptualization, metadata changes, and temporal dimension. Our change specification is represented by a set of changes. A set of changes consists of instance data change, structural change, and identifier change. In order to support a query in ontology versions, we consider temporal dimension includes valid time. Ontology versioning brings about massive amount of versions to be stored and maintained. We present the ontology versions management schemes that are 1) storing all the change sets, 2) storing the aggregation of change sets periodically, and 3) storing the aggregation of change sets using an adaptive criterion. We conduct a set of experiments to compare the performance of each versions management schemes. We present the experimental results for evaluating the performance of the three version management schemes from scheme 1 to scheme 3. Scheme 1 has the least storage usage. The average response time in Scheme 1 is extremely large, those of Scheme 3 is smaller than Scheme 2. Scheme 3 shows a good performance relatively.

• Journal of the Korean Society of Clothing and Textiles
• /
• v.32 no.2
• /
• pp.201-211
• /
• 2008

The purpose of this study was to clarify the concept of consideration set of stores and to contribute to the prediction of consumers' store behavior by finding out which consumer characteristics affect the formation of consideration set of stores. The data were collected from 553 female consumers. Factor analysis, K-cluster analysis, and ANOVA were used for data analysis. The results of this study were as follows: First, the respondents were classified into seven groups based on the number of stores and store types they considered. Second, there were significant differences among groups in consumer characteristics such as knowledge, motivation, and solubility; especially, the differences were related to the number of considering store rather than the types of considering store. In other words, the more involvement in clothing and the more experience and knowledge about apparel purchases a consumer had, the more stores the consumer considered. This study is meaningful in that it provides a systematic organization of the concept of consideration set of stores.