• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.11
• /
• pp.473-482
• /
• 2021

Although several Hangul generation models using deep learning have been introduced, they require a lot of data, have a complex structure, requires considerable time and resources, and often fail in style conversion. This paper proposes a model CKFont using the components of the initial, middle, and final components of Hangul as a way to compensate for these problems. The CKFont model is an end-to-end Hangul generation model based on GAN, and it can generate all Hangul in various styles with 28 characters and components of first, middle, and final components of Hangul characters. By acquiring local style information from components, the information is more accurate than global information acquisition, and the result of style conversion improves as it can reduce information loss. This is a model that uses the minimum number of characters among known models, and it is an efficient model that reduces style conversion failures, has a concise structure, and saves time and resources. The concept using components can be used for various image transformations and compositing as well as transformations of other languages.

• The Korean Journal of Applied Statistics
• /
• v.36 no.2
• /
• pp.129-139
• /
• 2023

Forecasting volatility is essential to avoiding the risk caused by the uncertainties of an financial asset. Complicated financial volatility features such as ambiguity between non-stationarity and stationarity, asymmetry, long-memory, sudden fairly large values like outliers bring great challenges to volatility forecasts. In order to address such complicated features implicity, we consider machine leaning models such as LSTM (1997) and GRU (2014), which are known to be suitable for existing time series forecasting. However, there are the problems of vanishing gradients, of enormous amount of computation, and of a huge memory. To solve these problems, a causal temporal convolutional network (TCN) model, an advanced form of 1D CNN, is also applied. It is confirmed that the overall forecasting power of TCN model is higher than that of the RNN models in forecasting VIX, VXD, and VXN, the daily volatility indices of S&P 500, DJIA, Nasdaq, respectively.

• Smart Media Journal
• /
• v.12 no.7
• /
• pp.76-83
• /
• 2023

This study proposes a method to assist webtoon artists in the process of webtoon creation by utilizing a pretrained Text-to-Image model to generate webtoon images from text. The proposed approach involves fine-tuning a pretrained Stable Diffusion model using a webtoon dataset transformed into the desired webtoon style. The fine-tuning process, using LoRA technique, completes in a quick training time of approximately 4.5 hours with 30,000 steps. The generated images exhibit the representation of shapes and backgrounds based on the input text, resulting in the creation of webtoon-like images. Furthermore, the quantitative evaluation using the Inception score shows that the proposed method outperforms DCGAN-based Text-to-Image models. If webtoon artists adopt the proposed Text-to-Image model for webtoon creation, it is expected to significantly reduce the time required for the creative process.

• Journal of the Korea Society of Computer and Information
• /
• v.28 no.8
• /
• pp.67-75
• /
• 2023

Stock price prediction is an important topic extensively discussed in the financial market, but it is considered a challenging subject due to numerous factors that can influence it. In this research, performance was compared and analyzed by applying time series prediction models (LSTM, GRU) and non-time series prediction models (RF, SVR, KNN, LGBM) that do not take into account the temporal dependence of data into stock price prediction. In addition, various data such as stock price data, technical indicators, financial statements indicators, buy sell indicators, short selling, and foreign indicators were combined to find optimal predictors and analyze major factors affecting stock price prediction by industry. Through the hyperparameter optimization process, the process of improving the prediction performance for each algorithm was also conducted to analyze the factors affecting the performance. As a result of feature selection and hyperparameter optimization, it was found that the forecast accuracy of the time series prediction algorithm GRU and LSTM+GRU was the highest.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2021.10a
• /
• pp.297-299
• /
• 2021

Predicting future productivity of tight oil is an important task for analyzing residual oil recovery and reservoir behavior. In general, productivity prediction is made using the decline curve analysis(DCA). In this study, we intend to propose an effective model for predicting future production using deep learning-based recurrent neural networks(RNN), LSTM, and GRU algorithms. As input variables, the main parameters are oil, gas, water, which are calculated during the production of tight oil, and the type curve calculated through various cluster analyzes. the output variable is the monthly oil production. Existing empirical models, the DCA and RNN models, were compared, and an optimal model was derived through hyperparameter tuning to improve the predictive performance of the model.

• Steel and Composite Structures
• /
• v.49 no.2
• /
• pp.161-180
• /
• 2023

In the previous research, the axial compressive capacity models for the glass fiber-reinforced polymer (GFRP)-reinforced circular concrete compression elements restrained with GFRP helix were put forward based on small and noisy datasets by considering a limited number of parameters portraying less accuracy. Consequently, it is important to recommend an accurate model based on a refined and large testing dataset that considers various parameters of such components. The core objective and novelty of the current research is to suggest a deep learning model for the axial compressive capacity of GFRP-reinforced circular concrete columns restrained with a GFRP helix utilizing various parameters of a large experimental dataset to give the maximum precision of the estimates. To achieve this aim, a test dataset of 61 GFRP-reinforced circular concrete columns restrained with a GFRP helix has been created from prior studies. An assessment of 15 diverse theoretical models is carried out utilizing different statistical coefficients over the created dataset. A novel model utilizing the group method of data handling (GMDH) has been put forward. The recommended model depicted good effectiveness over the created dataset by assuming the axial involvement of GFRP main bars and the confining effectiveness of transverse GFRP helix and depicted the maximum precision with MAE = 195.67, RMSE = 255.41, and R2 = 0.94 as associated with the previously recommended equations. The GMDH model also depicted good effectiveness for the normal distribution of estimates with only a 2.5% discrepancy from unity. The recommended model can accurately calculate the axial compressive capacity of FRP-reinforced concrete compression elements that can be considered for further analysis and design of such components in the field of structural engineering.

• The Transactions of the Korea Information Processing Society
• /
• v.13 no.3
• /
• pp.140-147
• /
• 2024

In the Republic of Korea, people with hearing impairments are the second-largest demographic within the registered disability community, following those with physical disabilities. Despite this demographic significance, research on sign language translation technology is limited due to several reasons including the limited market size and the lack of adequately annotated datasets. Despite the difficulties, a few researchers continue to improve the performacne of sign language translation technologies by employing the recent advance of deep learning, for example, the transformer architecture, as the transformer-based models have demonstrated noteworthy performance in tasks such as action recognition and video classification. This study focuses on enhancing the recognition performance of sign language translation by combining transformers with 3D-CNN. Through experimental evaluations using the PHOENIX-Wether-2014T dataset [1], we show that the proposed model exhibits comparable performance to existing models in terms of Floating Point Operations Per Second (FLOPs).

• Journal of Intelligence and Information Systems
• /
• v.23 no.3
• /
• pp.139-153
• /
• 2017

Corporate bankruptcy can cause great losses not only to stakeholders but also to many related sectors in society. Through the economic crises, bankruptcy have increased and bankruptcy prediction models have become more and more important. Therefore, corporate bankruptcy has been regarded as one of the major topics of research in business management. Also, many studies in the industry are in progress and important. Previous studies attempted to utilize various methodologies to improve the bankruptcy prediction accuracy and to resolve the overfitting problem, such as Multivariate Discriminant Analysis (MDA), Generalized Linear Model (GLM). These methods are based on statistics. Recently, researchers have used machine learning methodologies such as Support Vector Machine (SVM), Artificial Neural Network (ANN). Furthermore, fuzzy theory and genetic algorithms were used. Because of this change, many of bankruptcy models are developed. Also, performance has been improved. In general, the company's financial and accounting information will change over time. Likewise, the market situation also changes, so there are many difficulties in predicting bankruptcy only with information at a certain point in time. However, even though traditional research has problems that don't take into account the time effect, dynamic model has not been studied much. When we ignore the time effect, we get the biased results. So the static model may not be suitable for predicting bankruptcy. Thus, using the dynamic model, there is a possibility that bankruptcy prediction model is improved. In this paper, we propose RNN (Recurrent Neural Network) which is one of the deep learning methodologies. The RNN learns time series data and the performance is known to be good. Prior to experiment, we selected non-financial firms listed on the KOSPI, KOSDAQ and KONEX markets from 2010 to 2016 for the estimation of the bankruptcy prediction model and the comparison of forecasting performance. In order to prevent a mistake of predicting bankruptcy by using the financial information already reflected in the deterioration of the financial condition of the company, the financial information was collected with a lag of two years, and the default period was defined from January to December of the year. Then we defined the bankruptcy. The bankruptcy we defined is the abolition of the listing due to sluggish earnings. We confirmed abolition of the list at KIND that is corporate stock information website. Then we selected variables at previous papers. The first set of variables are Z-score variables. These variables have become traditional variables in predicting bankruptcy. The second set of variables are dynamic variable set. Finally we selected 240 normal companies and 226 bankrupt companies at the first variable set. Likewise, we selected 229 normal companies and 226 bankrupt companies at the second variable set. We created a model that reflects dynamic changes in time-series financial data and by comparing the suggested model with the analysis of existing bankruptcy predictive models, we found that the suggested model could help to improve the accuracy of bankruptcy predictions. We used financial data in KIS Value (Financial database) and selected Multivariate Discriminant Analysis (MDA), Generalized Linear Model called logistic regression (GLM), Support Vector Machine (SVM), Artificial Neural Network (ANN) model as benchmark. The result of the experiment proved that RNN's performance was better than comparative model. The accuracy of RNN was high in both sets of variables and the Area Under the Curve (AUC) value was also high. Also when we saw the hit-ratio table, the ratio of RNNs that predicted a poor company to be bankrupt was higher than that of other comparative models. However the limitation of this paper is that an overfitting problem occurs during RNN learning. But we expect to be able to solve the overfitting problem by selecting more learning data and appropriate variables. From these result, it is expected that this research will contribute to the development of a bankruptcy prediction by proposing a new dynamic model.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.34 no.3
• /
• pp.439-449
• /
• 2024

Recent rapid advancements in voice generation technology have enabled the natural synthesis of voices using text alone. However, this progress has led to an increase in malicious activities, such as voice phishing (voishing), where generated voices are exploited for criminal purposes. Numerous models have been developed to detect the presence of synthesized voices, typically by extracting features from the voice and using these features to determine the likelihood of voice generation.This paper proposes a new model for extracting voice features to address misuse cases arising from generated voices. It utilizes a deep learning-based audio codec model and the pre-trained natural language processing model BERT to extract novel voice features. To assess the suitability of the proposed voice feature extraction model for voice detection, four generated voice detection models were created using the extracted features, and performance evaluations were conducted. For performance comparison, three voice detection models based on Deepfeature proposed in previous studies were evaluated against other models in terms of accuracy and EER. The model proposed in this paper achieved an accuracy of 88.08%and a low EER of 11.79%, outperforming the existing models. These results confirm that the voice feature extraction method introduced in this paper can be an effective tool for distinguishing between generated and real voices.

• Korean Journal of Remote Sensing
• /
• v.38 no.6_1
• /
• pp.1505-1514
• /
• 2022

Image matching is a crucial preprocessing step for effective utilization of multi-temporal and multi-sensor very high resolution (VHR) satellite images. Deep learning (DL) method which is attracting widespread interest has proven to be an efficient approach to measure the similarity between image pairs in quick and accurate manner by extracting complex and detailed features from satellite images. However, Image matching of VHR satellite images remains challenging due to limitations of DL models in which the results are depending on the quantity and quality of training dataset, as well as the difficulty of creating training dataset with VHR satellite images. Therefore, this study examines the feasibility of DL-based method in matching pair extraction which is the most time-consuming process during image registration. This paper also aims to analyze factors that affect the accuracy based on the configuration of training dataset, when developing training dataset from existing multi-sensor VHR image database with bias for DL-based image matching. For this purpose, the generated training dataset were composed of correct matching pairs and incorrect matching pairs by assigning true and false labels to image pairs extracted using a grid-based Scale Invariant Feature Transform (SIFT) algorithm for a total of 12 multi-temporal and multi-sensor VHR images. The Siamese convolutional neural network (SCNN), proposed for matching pair extraction on constructed training dataset, proceeds with model learning and measures similarities by passing two images in parallel to the two identical convolutional neural network structures. The results from this study confirm that data acquired from VHR satellite image database can be used as DL training dataset and indicate the potential to improve efficiency of the matching process by appropriate configuration of multi-sensor images. DL-based image matching techniques using multi-sensor VHR satellite images are expected to replace existing manual-based feature extraction methods based on its stable performance, thus further develop into an integrated DL-based image registration framework.