• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.11a
• /
• pp.127-128
• /
• 2020

본 논문은 딥러닝 네트워크의 압축을 위한 양자화 오프셋의 바이어스 기법을 제안한다. 양자화는 32비트 정밀도를 갖는 가중치와 활성화 데이터를 특정 비트 이하의 정수로 압축한다. 양자화는 원 데이터에 스케일과 오프셋을 더함으로써 수행되므로 오프셋을 위한 합성곱 연산이 추가된다. 본 논문에서는 입력 활성화 데이터의 양자화 오프셋과 가중치의 합성곱의 출력은 바이어스에 임베딩될 수 있음을 보여준다. 이를 통해 추론 과정 중 오프셋의 합성곱 연산을 제거할 수 있다. 실험 결과는 오프셋의 합성곱이 바이어스에 임베딩이 되더라도 영상 분류 정확도에 영향이 거의 없음을 증명한다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.11a
• /
• pp.25-28
• /
• 2020

Automatic segmentation of brain tissues such as WM, GM, and CSF from brain MRI scans is helpful for the diagnosis of many neurological disorders. Accurate segmentation of these brain structures is a very challenging task due to low tissue contrast, bias filed, and partial volume effects. With the aim to improve brain MRI segmentation accuracy, we propose an end-to-end convolutional based U-SegNet architecture designed with multi-scale kernels, which includes cascaded dilated convolutions for the task of brain MRI segmentation. The multi-scale convolution kernels are designed to extract abundant semantic features and capture context information at different scales. Further, the cascaded dilated convolution scheme helps to alleviate the vanishing gradient problem in the proposed model. Experimental outcomes indicate that the proposed architecture is superior to the traditional deep-learning methods such as Segnet, U-net, and U-Segnet and achieves high performance with an average DSC of 93% and 86% of JI value for brain MRI segmentation.

• Analytical Science and Technology
• /
• v.12 no.3
• /
• pp.177-183
• /
• 1999

The determination method of free acid in spent U-Al nuclear fuel solutions by Gran plot titration was described. Effect of U(VI) and Al(III) on the alkalimetric titration of nitric acid was investigated in oxalate complexing media as well as in noncomplexing media. Positive biases were observed in both titration media when the end-point was estimated by the Gran plot method. It was found that the cause of the bias was U(VI) in the oxalate complexing media, but Al(III) in the noncomplexing media. The relative error was less than 1% in the titration of 0.1 M $HNO_3$ at a U(VI) : Al(III) : $H^+$ mole ratio of up to 2:12:1 as long as the pH of the oxalate titration media was sustained to be below 5.0 at the beginning of titration. The method was successfully applied to the determination of nitric acid in a solution of HANARO reactor fuel with U:Al mole ratio of 1:6.

• Journal of the Korean Magnetics Society
• /
• v.10 no.6
• /
• pp.310-317
• /
• 2000

With use of (G)MR head, thermal asperity (TA) has been a big concern in drive industry. In this study, we investigated several factors of heads and disks which affects the TA sensitivity of the drive. TA experiments were conducted by introducing the particles on the drives using a particle injection chamber. It was found that the slider ABS shape can help to reduce TA or contamination in the head/media interface. However, TA sensitivity of the drive mainly depend on the intrinsic property of (G)MR sensor. GMR head is much less sensitive to TA compared with MR head. However, in case that the same bias current was applied for both of MR and GMR head, TA sensitivity of GMR head became almost identical to that of MR head. Therefore it was found that the bias current is a dominant factor in determining TA sensitivity of the head. TA sensitivity of different types of disks was also studied. The scratch resistance of the carbon overcoat layer is the one of the main factors which influence TA rejection capability of the disks.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.7
• /
• pp.949-955
• /
• 2022

Deep learning-based text summarization models are not free from datasets. For example, a summarization model trained with a news summarization dataset is not good at summarizing other types of texts such as internet posts and papers. In this study, we define this phenomenon as Data Bias Problem (DBP) and propose two training methods for solving it. The first is the 'proper nouns masking' that masks proper nouns. The second is the 'length variation' that randomly inflates or deflates the length of text. As a result, experiments show that our methods are efficient for solving DBP. In addition, we analyze the results of the experiments and present future development directions. Our contributions are as follows: (1) We discovered DBP and defined it for the first time. (2) We proposed two efficient training methods and conducted actual experiments. (3) Our methods can be applied to all summarization models and are easy to implement, so highly practical.

• Nuclear Engineering and Technology
• /
• v.55 no.9
• /
• pp.3450-3463
• /
• 2023

In the two-step analysis of Pebble-Bed type High-Temperature Gas-Cooled Reactor (PB-HTGR), the lattice physics calculation for the generation of homogenized cross-sections is based on the fuel pebble. However, the randomly-dispersed fuel particles in the fuel pebble introduce double heterogeneity and randomness. Compared to the deterministic method, the Monte Carlo method which is flexible in geometry modeling provides a high-fidelity treatment. Therefore, the Monte Carlo code NECP-MCX is extended in this study to perform the lattice physics calculation of the PB-HTGR. Firstly, the capability for the simulation of randomly-dispersed media, using the explicit modeling approach, is developed in NECP-MCX. Secondly, the capability for the generation of the homogenized cross-section is also developed in NECP-MCX. Finally, simplified PB-HTGR problems are calculated by a two-step neutronics analysis tool based on Monte Carlo homogenization. For the pebble beds mixed by fuel pebble and graphite pebble, the bias is less than 100 pcm when compared to the high-fidelity model, and the bias is increased to 269 pcm for pebble bed mixed by depleted fuel pebble. Numerical results show that the Monte Carlo lattice physics calculation for the two-step analysis of PB-HTGR is feasible.

• Asia pacific journal of information systems
• /
• v.24 no.2
• /
• pp.191-210
• /
• 2014

As mobile devices that can be connected to the Internet have spread and networking has become possible whenever/wherever, the Internet has become central in the dissemination and consumption of news. Accordingly, the ways news is gathered, disseminated, and consumed have changed greatly. In the traditional news media such as magazines and newspapers, expert editors determined what events were worthy of deploying their staffs or freelancers to cover and what stories from newswires or other sources would be printed. Furthermore, they determined how these stories would be displayed in their publications in terms of page placement, space allocation, type sizes, photographs, and other graphic elements. In turn, readers-news consumers-judged the importance of news not only by its subject and content, but also through subsidiary information such as its location and how it was displayed. Their judgments reflected their acceptance of an assumption that these expert editors had the knowledge and ability not only to serve as gatekeepers in determining what news was valuable and important but also how to rank its value and importance. As such, news assembled, dispensed, and consumed in this manner can be said to be expert-based recommended news. However, in the era of Internet news, the role of expert editors as gatekeepers has been greatly diminished. Many Internet news sites offer a huge volume of news on diverse topics from many media companies, thereby eliminating in many cases the gatekeeper role of expert editors. One result has been to turn news users from passive receptacles into activists who search for news that reflects their interests or tastes. To solve the problem of an overload of information and enhance the efficiency of news users' searches, Internet news sites have introduced numerous recommendation techniques. Recommendations based on popularity constitute one of the most frequently used of these techniques. This popularity-based approach shows a list of those news items that have been read and shared by many people, based on users' behavior such as clicks, evaluations, and sharing. "most-viewed list," "most-replied list," and "real-time issue" found on news sites belong to this system. Given that collective intelligence serves as the premise of these popularity-based recommendations, popularity-based news recommendations would be considered highly important because stories that have been read and shared by many people are presumably more likely to be better than those preferred by only a few people. However, these recommendations may reflect a popularity bias because stories judged likely to be more popular have been placed where they will be most noticeable. As a result, such stories are more likely to be continuously exposed and included in popularity-based recommended news lists. Popular news stories cannot be said to be necessarily those that are most important to readers. Given that many people use popularity-based recommended news and that the popularity-based recommendation approach greatly affects patterns of news use, a review of whether popularity-based news recommendations actually reflect important news can be said to be an indispensable procedure. Therefore, in this study, popularity-based news recommendations of an Internet news portal was compared with top placements of news in printed newspapers, and news users' judgments of which stories were personally and socially important were analyzed. The study was conducted in two stages. In the first stage, content analyses were used to compare the content of the popularity-based news recommendations of an Internet news site with those of the expert-based news recommendations of printed newspapers. Five days of news stories were collected. "most-viewed list" of the Naver portal site were used as the popularity-based recommendations; the expert-based recommendations were represented by the top pieces of news from five major daily newspapers-the Chosun Ilbo, the JoongAng Ilbo, the Dong-A Daily News, the Hankyoreh Shinmun, and the Kyunghyang Shinmun. In the second stage, along with the news stories collected in the first stage, some Internet news stories and some news stories from printed newspapers that the Internet and the newspapers did not have in common were randomly extracted and used in online questionnaire surveys that asked the importance of these selected news stories. According to our analysis, only 10.81% of the popularity-based news recommendations were similar in content with the expert-based news judgments. Therefore, the content of popularity-based news recommendations appears to be quite different from the content of expert-based recommendations. The differences in importance between these two groups of news stories were analyzed, and the results indicated that whereas the two groups did not differ significantly in their recommendations of stories of personal importance, the expert-based recommendations ranked higher in social importance. This study has importance for theory in its examination of popularity-based news recommendations from the two theoretical viewpoints of collective intelligence and popularity bias and by its use of both qualitative (content analysis) and quantitative methods (questionnaires). It also sheds light on the differences in the role of media channels that fulfill an agenda-setting function and Internet news sites that treat news from the viewpoint of markets.

• The Journal of the Korea Contents Association
• /
• v.21 no.6
• /
• pp.466-479
• /
• 2021

Fake news is flooded with fake news that mixes untrue falsehoods in whole or in part. In order to create a frame for political and social purposes, news is mainly created by combining facts with fakes, or news is created for the purpose of incitement to encourage distortion and hatred. In particular, some media, including over-the-air broadcasting, are expressing their political bias in a specific direction without hesitation. Even if the press has political bias, if it is based on the delivery of facts, a trusted media environment can be achieved under the mutual checks of conservatives and progressives. The May 18 Democratic Uprising is a painful history and fact of modern history, leaving remuneration and progress. Nevertheless, there is still a view that the May 18 Democratic Uprising is regarded as 'democracy movement' and 'violence'. This study analyzed how the conservative JoongAng Ilbo and the progressive Kyunghyang Shinmun Manpyeong deal with the May 18 Democratic Uprising. The Manpyeong of the two newspapers differs greatly from the viewpoint of the May 18 Democratic Uprising. The liberal tendency of the Kyunghyang Shinmun Manpyeong has great significance in that it reveals the essence of the event and satirizes the subject. On the other hand, the conservative JoongAng Ilbo Manpyeong cannot approach the nature of the case or the object of satire due to ambiguity.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.22 no.3
• /
• pp.25-30
• /
• 2022

This study is a study on the filter bubble reinforcement phenomenon of SNS recommendation algorithm such as YouTube, which is a characteristic of the Russian-Ukraine war (2022), and the victory or defeat factors of the hybrid war. This war is identified as a hybrid war, and the use of New Media based on the SNS recommendation algorithm is emerging as a factor that determines the outcome of the war beyond political leverage. For this reason, the filter bubble phenomenon goes beyond the dictionary meaning of confirmation bias that limits information exposed to viewers. A YouTube video of Ukrainian President Zelensky encouraging protests in Kyiv garnered 7.02 million views, but Putin's speech only 800,000, which is a evidence that his speech was not exposed to the recommendation algorithm. The war of these SNS recommendation algorithms tends to develop into an algorithm war between the US (YouTube, Twitter, Facebook) and China (TikTok) big tech companies. Influenced by US companies, Ukraine is now able to receive international support, and in Russia, under the influence of Chinese companies, Putin's approval rating is over 80%, resulting in conflicting results. Since this algorithmic empowerment is based on the confirmation bias of public opinion by 'filter bubble', the justification that a new guideline setting for this distortion phenomenon should be presented shortly is drawing attention through this Russia-Ukraine war.

• The Korea Journal of Herbology
• /
• v.36 no.5
• /
• pp.1-14
• /
• 2021

Objectives : From January 2020, when the first confirmed case of COVID-19 in Korea, the use of health information using the Internet is expected to increase. It is expected that there will be a significant change in the general public's interest in Korean herbal medicines for health care. Therefore, in this study, we tried to confirm the change in the search trend of Korean herbal medicines after the COVID-19 epidemic. Methods : Using the "Naver DataLab (http://datalab.naver.com)" service of a Korean portal site Naver, search volume was investigated with 606 Korean herbal medicines as keywords. The search period was from January 2020, right after the onset of COVID-19, to June 2021. The search results were sorted by the peak search volume and the total search volume. Results : 'Cheonsangap (천산갑, 穿山甲, Manitis Squama)' was the most searched Korean herbal medicine in the peak search volume and total search volume with least bias. Conclusions : The problem of supply and demand of Korean herbal medicines of high public interest was identified. Broadcasting and media exposure were the factors that had a big impact on the search volume for Korean herbal medicines. As it was confirmed that the search volume for Korean herbal medicines increased rapidly due to media exposure, it is necessary to provide correct information about Korean herbal medicines, improve public awareness, and manage stable supply and demand based on continuous search trend monitoring.