• Journal of the Korea Society of Computer and Information
• /
• v.26 no.10
• /
• pp.157-165
• /
• 2021

Recently, studies using deep learning to analyze a large amount of text are being actively conducted. In particular, a pre-trained language model that applies the learning results of a large amount of text to the analysis of a specific domain text is attracting attention. Among various pre-trained language models, BERT(Bidirectional Encoder Representations from Transformers)-based model is the most widely used. Recently, research to improve the performance of analysis is being conducted through further pre-training using BERT's MLM(Masked Language Model). However, the traditional MLM has difficulties in clearly understands the meaning of sentences containing new words such as newly coined words. Therefore, in this study, we newly propose NTM(Newly coined words Target Masking), which performs masking only on new words. As a result of analyzing about 700,000 movie reviews of portal 'N' by applying the proposed methodology, it was confirmed that the proposed NTM showed superior performance in terms of accuracy of sensitivity analysis compared to the existing random masking.

• Electrical & Electronic Materials
• /
• v.9 no.5
• /
• pp.490-497
• /
• 1996

RF마그네트론 스파터링에 의한 $sSnO_2$ 박막을 증착한 결과 작동조건에 따라 표면에서 막손상이 발생하였는데, 이는 target에서 생성된 높은 에너지의 고속입자가 기판에서 직접 충돌로 일어나며, 마그네트론의 자계분포와도 관계된다. 또한 기판의 위치에 따라 박막의 전기적, 구조적 특성이 급격히 변하는데 본 논문은 중심자석의 세기와 RF power, 가스압력, 그리고 기판온도 등의 스파터링 작동조건을 변화시키면서 박막의 비손상을 검토하였고, 물성특성을 평가하였다. 실험결과로 부터 기판의 외축부에서 제작된 박막은 전반적으로 막손상이 없었고, 특히 target의 중심자석을 Cobalt로 설치하고 15 mTorr의 가스압력과 50 W의 RF power로 한 경우 가장 우수한 특성을 가진 박막을 얻을 수 있었다. 추가로 막손상을 방지하고자, 환원형의 masking glass를 target위에 설치하였는데, 고에너지 입자의 직접 충돌을 확실히 차단할 수 있었으며, 비저항율도 target의 부식부위 (erosion)에 대응하는 부분에서 100배 정도로 개선하였다.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.49 no.9
• /
• pp.55-64
• /
• 2012

Until now, Side Channel Attack has been known to be effective to crack decrypt key such as smart cards, electronic passports and e-ID card based on Chip. Combination of Masking and shuffling methods have been proposed practical countermeasure. Newly, S.Tillich suggests biased-mask using template attack(TA) to attack AES with masking and shuffling. However, an additional assumption that is acquired template information previously for masking value is necessary in order to apply this method. Moreover, this method needs to know exact time position of the target masking value for higher probability of success. In this paper, we suggest new practical method called Biasing Power Analysis(BPA) to find a secret key of AES based on masking-shuffling method. In BPA, we don't use time position and template information from masking value. Actually, we do experimental works of BPA attack to 128bit secret key of AES based on masking-shuffling method performed MSP430 Chip and we succeed in finding whole secret key. The results of this study will be utilized for next-generation ID cards to verify physical safety.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2021.07a
• /
• pp.391-394
• /
• 2021

최근 딥러닝(Deep Learning)을 활용하여 텍스트로 표현된 단어나 문장의 의미를 파악하기 위한 다양한 연구가 활발하게 수행되고 있다. 하지만, 딥러닝을 통해 특정 도메인에서 사용되는 언어를 이해하기 위해서는 해당 도메인의 충분한 데이터에 대해 오랜 시간 학습이 수행되어야 한다는 어려움이 있다. 이러한 어려움을 극복하고자, 최근에는 방대한 양의 데이터에 대한 학습 결과인 사전 학습 언어 모델(Pre-trained Language Model)을 다른 도메인의 학습에 적용하는 방법이 딥러닝 연구에서 많이 사용되고 있다. 이들 접근법은 사전 학습을 통해 단어의 일반적인 의미를 학습하고, 이후에 단어가 특정 도메인에서 갖는 의미를 파악하기 위해 추가적인 학습을 진행한다. 추가 학습에는 일반적으로 대표적인 사전 학습 언어 모델인 BERT의 MLM(Masked Language Model)이 다시 사용되며, 마스크(Mask) 되지 않은 단어들의 의미로부터 마스크 된 단어의 의미를 추론하는 형태로 학습이 이루어진다. 따라서 사전 학습을 통해 의미가 파악되어 있는 단어들이 마스크 되지 않고, 신조어와 같이 의미가 알려져 있지 않은 단어들이 마스크 되는 비율이 높을수록 단어 의미의 학습이 정확하게 이루어지게 된다. 하지만 기존의 MLM은 무작위로 마스크 대상 단어를 선정하므로, 사전 학습을 통해 의미가 파악된 단어와 사전 학습에 포함되지 않아 의미 파악이 이루어지지 않은 신조어가 별도의 구분 없이 마스크에 포함된다. 따라서 본 연구에서는 사전 학습에 포함되지 않았던 신조어에 대해서만 집중적으로 마스킹(Masking)을 수행하는 방안을 제시한다. 이를 통해 신조어의 의미 학습이 더욱 정확하게 이루어질 수 있고, 궁극적으로 이러한 학습 결과를 활용한 후속 분석의 품질도 향상시킬 수 있을 것으로 기대한다. 영화 정보 제공 사이트인 N사로부터 영화 댓글 12만 건을 수집하여 실험을 수행한 결과, 제안하는 신조어 표적 마스킹(NTM: Newly Coined Words Target Masking)이 기존의 무작위 마스킹에 비해 감성 분석의 정확도 측면에서 우수한 성능을 보임을 확인하였다.

• Journal of Advanced Navigation Technology
• /
• v.2 no.1
• /
• pp.43-52
• /
• 1998

In this paper, we have investigated the effects of geographic information on the performance of multiple ground target tracking system using multiple sensors. Geographic information is utilized in two cases: association and masking target measurement. Virtually no improvement is observed to the overall performance of tracking system when we applied mobility to the association procedure. Masking target measurement based on mobility produces desirable result that the number of false tracks is reduced. Since geographic information can be regarded as an additional sensor in sensor fusion paradigm, careful usage is required.

• Proceedings of the KSRS Conference
• /
• 2005.10a
• /
• pp.120-122
• /
• 2005

With the commercial availability of very high resolution satellite imagery, the concealment of national confidential targets such as military facilities became one of the most bothering task to the image distributors. This task has been carried out by handwork masking of the target objects. Therefore, the quality of the concealment was fully depends on the ability and skill of a worker. In this study, a spectral clustering based technique for the seamless concealment of confidential targets in high resolution imagery was developed. The applicability test shows that the proposed technique can be used as a practical procedure for those who need to hide some information in image before public distribution

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.18 no.12
• /
• pp.2961-2967
• /
• 2014

Nowadays, a private protection act has come into effect which demands for the protection of personal image information obtained by the CCTV. According to this act, the object out of interest has to be mosaicked such that it can not be identified before the image is sent to the investigation office. Meanwhile, the demand for digital videos obtained by CCTV is also increasing for digital forensic. Therefore, due to the two conflicting demands, the demand for a solution which can automatically mask an object in the CCTV video is increasing and related IT industry is expected to grow. The core technology in developing a target masking solution is the object tracking technique. In this paper, we propose an object tracking technique which suits for the application of CCTV video object masking as a postprocess. The proposed method simultaneously uses the motion and the color information to produce a stable tracking result. Furthermore, the proposed method is based on the centroid shifting method, which is a fast color based tracking method, and thus the overall tracking becomes fast.

• Journal of Institute of Control, Robotics and Systems
• /
• v.6 no.2
• /
• pp.173-180
• /
• 2000

During the last decade many researches have been working on multiple target tracking problem in the area of radar application, Various approaches have been proposed to solve the tracking problem and the concept of sensor fusion was established as an effort. In this paper utilization of geographic information for ground target tracking is investigated and performance comparison with the results of applying sensor fusion is described. Geographic information is used in three aspects: association masking target measurement and re-striction of removing true target. Simulation results indicate that using two sensors shows better performance with respect to tracking but a single with geographic information is a winner in reducing the number of false tracks.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.17 no.1
• /
• pp.32-40
• /
• 2013

In this paper, we propose a multi-channel speech enhancement algorithm using beamforming and direction-of-arrival (DOA)-based masking. The proposed algorithm enhances noisy speech basically by the linearly constrained minimum variance (LCMV) algorithm and then a mel-scale Wiener filter designed using DOA-based masking is applied to remove still remaining noises. To improve the performance, we optimize the learning rate of the adaptive filters in LCMV and the DOA threshold to detect target speech spectrum. As performance indices, the perceptual evaluation of speech quality (PESQ) score and output SNRs are measured. Experimantal results show that the proposed algorithm outperforms the conventional LCMV beamformer by 0.09 in PESQ score and 5.75 dB in output SNR, respectively.

• Journal of Audiology & Otology
• /
• v.23 no.1
• /
• pp.27-32
• /
• 2019

Background and Objectives: Detection of auditory signals may be improved when maskers far from the frequency of the target signal are coherently amplitude-modulated. This improvement of signal detection is called comodulation masking release (CMR). In the CMR experiments, flankers have been usually arranged symmetrically. In practice, we will be confronted with a problem by using symmetric flankers due to the limited output of clinical audiometers, especially at high-frequency. We aimed to check whether flanker arrangement has any effect on the amount of CMR, especially when there is no flankers with a frequency higher than the signal. Subjects and Methods: Eighteen normal hearing listeners ranging in age from 20 to 46 years old participated. Symmetric (2-2) and asymmetric (3-1 and 4-0) flankers were used and then the amount of CMR compared among them. Results: Our results showed in the same numbers of flankers, there were no statistically CMR differences between symmetric and asymmetric arrangement. Also when we did not have a flanker at a frequency higher than the signal and all flankers were placed below the signal, there was no statistically difference with the symmetric arrangement. Conclusions: The asymmetry of the flankers and also omitting the flankers with a frequency higher than the signal, have no effect on CMR results. We concluded that CMR can be considered by using clinical audiometer.