• Journal of Information Technology Services
• /
• v.19 no.5
• /
• pp.83-91
• /
• 2020

Language Models such as BERT has been an important factor of deep learning-based natural language processing. Pre-training the transformer-based language models would be computationally expensive since they are consist of deep and broad architecture and layers using an attention mechanism and also require huge amount of data to train. Hence, it became mandatory to do fine-tuning large pre-trained language models which are trained by Google or some companies can afford the resources and cost. There are various techniques for fine tuning the language models and this paper examines three techniques, which are data augmentation, tuning the hyper paramters and partly re-constructing the neural networks. For data augmentation, we use no-answer augmentation and back-translation method. Also, some useful combinations of hyper parameters are observed by conducting a number of experiments. Finally, we have GRU, LSTM networks to boost our model performance with adding those networks to BERT pre-trained model. We do fine-tuning the pre-trained korean-based language model through the methods mentioned above and push the F1 score from baseline up to 89.66. Moreover, some failure attempts give us important lessons and tell us the further direction in a good way.

• Journal of Korea Multimedia Society
• /
• v.16 no.10
• /
• pp.1204-1213
• /
• 2013

With the development of digital satellite communication technology and the widespread use of smart devices, the demand for data communication in the maritime ship has increased. Recently, the communication between the maritime ship and the land is based on Inmarsat satellite service. The Inmarsat provides telephone, fax, data and telex service etc. However, since the satellite is payper-seconds billing service, the transmission of whole web contents to the maritime ship through the satellite causes high cost. In this paper, we propose web browsing system architecture that reduces the data traffic on the satellite link and retransmits the content selectively in order to solve these problems.

• Journal of KIISE:Software and Applications
• /
• v.32 no.9
• /
• pp.912-919
• /
• 2005

When we apply binary classification to multi-class classification for text categorization, we use the One-Against-All method generally, However, this One-Against-All method has a problem. That is, documents of a negative set are not labeled by human. Thus, they can include many noisy documents in the training data. In this paper, we propose that the Sliding Window technique and the EM algorithm are applied to binary text classification for solving this problem. We here improve binary text classification through extracting noise documents from the training data by the Sliding Window technique and re-assigning categories of these documents using the EM algorithm.

• 제어로봇시스템학회:학술대회논문집
• /
• 2003.10a
• /
• pp.2126-2130
• /
• 2003

In the integrated systems, that is integrated digital TV(DTV) internet and home automation, like home server, is needed integration of digital TV video signal and computer graphic signal. The graphic signal is operating at the high speed and has time-divide-stream. So the re-request of data is not easy at the time of error detection. therefore EDAC algorithm is efficient. In this paper, we show a scheme, that is integration of graphic and dtv format signal for DTV monitor display. This paper also presents the efficiency error detection auto correction(EDAC) for conversion of graphics signal to DTV video signal. A presented EDAC algorithms use the modified hamming code for enhancing video quality and reliability. A EDAC algorithm of this paper can detect single error, double error, triple error and more error for preventing from incorrect correction. And it is not necessary an additional memory. In this paper The comparison between digital TV video signal and graphic signal, a EDAC algorithm and a design of conversion graphic signal to DTV signal with EDAC function in DTV system is described.

• Proceedings of the IEEK Conference
• /
• 2002.07a
• /
• pp.106-109
• /
• 2002

In the integrated systems, that is integrated digital TV(DTV) internet and home automation, like home server, is needed integration of digital TV video signal and computer graphic signal. The graphic signal is operating at the high speed and has time-divide-stream. So the re-request of data is not easy at the time of error detection. therefore EDAC algorithm is efficient. This paper presents the efficiency error detection auto correction(EDAC) for conversion of graphics signal to DTV video signal. A presented EDAC algorithms use the modified Hamming code for enhancing video quality and reliability. A EDAC algorithm of this paper can detect single error, double error, triple error and more error for preventing from incorrect correction. And it is not necessary an additional memory. In this paper The comparison between digital TV video signal and graphic signal, a EBAC algorithm and a design of conversion graphic signal to DTV signal with EDAC function is described.

• Journal of the Korean Society of Fisheries and Ocean Technology
• /
• v.30 no.4
• /
• pp.263-272
• /
• 1994

The hardware and the software of the prototype telemtry system to monitor the behavior of the fish are designed. This system consistes of five parts I. e. three omni-directional hydrophones, three ultrasonic receivers, a single board computer for the signal processing, two RF transceivers for the data communication, and a personnel computer. The sensitivty of the hydrophones is -170dB(re 1V/$\mu$Pa), the gain and the 3dB receiving bandwidth of the ultrasonic receivers are 115dB and 1500Hz respectively, and the sampling period is 33.3$\mu$sec in the signal processing part. The positioning error of the system using hyperbolic method is estimated to be less than 0.2m in case that the pinger locates inside of the baselines. The perfomance of the system considering a practical use was examined by numerical simulation and a water tank test of a pinger tracking experiment. In results, the system developed in prototype was confirmed that it could be useful for monitoring the behavior of fish in the limited water area.

• Journal of Fashion Business
• /
• v.14 no.1
• /
• pp.116-131
• /
• 2010

As our world is becoming more and more globalized, nations tend to turn their interests towards their unique legacy and traditional culture. This research is intended to re-illuminate the Korean beauty through the Lotus Pattern, a traditional factor, from the Chosun Dynasty, and at the same time, analyze how its peculiar representation and figural elements can be reflected in modern fashion designs. The results will be an essential factor in creating exclusive and original designs. Research method was theoretical research from documents and to refer to positive data on preserved relics, and research contents consists of analysis on the use of lotus patterns in artworks that were exhibited during 2000 to 2008 by Korean and foreign artists. Results showed that lotus patterns that were used during the Chosun Dynasty are categorized by shape, composition form, and structural form. Applying these results, a total of 8 works were designed and created.

• 한국신재생에너지학회:학술대회논문집
• /
• 2009.06a
• /
• pp.592-595
• /
• 2009

In recent years, geothermal heat pump (GHP) systems have become increasingly popular for heating and cooling in buildings. The Standing Column Well (SCW) method is one of the most efficient GHP system. Because it use groundwater for heat transfer material. In SCW systems, water is re-circulated between the well and the building (heat pump). It is only a short time since this method has been applied in domestic. So we have to refer to the developed countries' guides and manuals of SCW. In this paper, several design and construction points of SCW method are filed. We used real operation data of SCW system at Chong-Ju Univ. site for economical efficiency analysis. As a result, the payback period of Chong-Ju Univ. site is calculated at 7.23 years.

• Management Science and Financial Engineering
• /
• v.5 no.2
• /
• pp.43-53
• /
• 1999

In many situations, the system failures depend on the operating environmental conditions that vary on time, usually with periodical manners. We use nonhomogeneous Poisson processes whose rate functions exhibit cyclic behavior as well as a long-term evolutionary trend to model the stochastic process of the failures when the rate of occurrence of the failures varies periodically, for example from day to day or between seasons. In this study, we compare optimal policies under the nonho-mogeneous process with/without a cyclic component in the failure rate function. The analytical re-sults for various situations are presented along with numerical examples using simulated data.

• Journal of the Korean Society of Food Culture
• /
• v.13 no.1
• /
• pp.27-35
• /
• 1998

Consumption of edible oil at food service institutions in Inchon was surveyed to provide basic data for continuous education of dietitians. Manufacturing industry was the major food service institution(78.8%) in Inchon followed by schools, hospitals and social welfare service centers. Most dietitians were at the age of twenties and college graduates with professional careers of 1-5 years. Oil was purchased on the basis of its quality within 1-3% of total food costs four times a month. Proper frying temperature was determined by dropping salt or food coating materials into the oil. Soybean oil was the most frequently used and commercial frying mix powder and flour and eggs in water were the most common food coating materials. Fish and commercial frozen foods were the most frequently used materials for frying. Fried foods were stored with covers in a basket and consumed within 30 minutes after cooking. Frying oils were used one more time after filtering and color was the index for determination of re-use.