• 한국어정보학회:학술대회논문집
• /
• 2016.10a
• /
• pp.300-304
• /
• 2016

본 논문에서는 한국어 의미역 결정을 순차열 분류 문제(Sequence Labeling Problem)가 아닌 순차열 변환 문제(Sequence-to-Sequence Learning)로 접근하였고, 구문 분석 단계와 자질 설계가 필요 없는 End-to-end 방식으로 연구를 진행하였다. 음절 단위의 RNN Search 모델을 사용하여 음절 단위로 입력된 문장을 의미역이 달린 어절들로 변환하였다. 또한 순차열 변환 문제의 성능을 높이기 위해 연구된 인풋-피딩(Input-feeding) 기술과 카피넷(CopyNet) 기술을 한국어 의미역 결정에 적용하였다. 실험 결과, Korean PropBank 데이터에서 79.42%의 레이블 단위 f1-score, 71.58%의 어절 단위 f1-score를 보였다.

• Annual Conference on Human and Language Technology
• /
• 2016.10a
• /
• pp.300-304
• /
• 2016

본 논문에서는 한국어 의미역 결정을 순차열 분류 문제(Sequence Labeling Problem)가 아닌 순차열 변환 문제(Sequence-to-Sequence Learning)로 접근하였고, 구문 분석 단계와 자질 설계가 필요 없는 End-to-end 방식으로 연구를 진행하였다. 음절 단위의 RNN Search 모델을 사용하여 음절 단위로 입력된 문장을 의미역이 달린 어절들로 변환하였다. 또한 순차열 변환 문제의 성능을 높이기 위해 연구된 인풋-피딩(Input-feeding) 기술과 카피넷(CopyNet) 기술을 한국어 의미역 결정에 적용하였다. 실험 결과, Korean PropBank 데이터에서 79.42%의 레이블 단위 f1-score, 71.58%의 어절 단위 f1-score를 보였다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2016.10a
• /
• pp.62-63
• /
• 2016

딥러닝의 대표적 개발 환경 중 하나인 Caffe를 임베디드 시스템의 메모리 구조를 고려하여 최적화하고 실제 측정 실험으로 기존의 방식보다 처리시간과 소비 전력량의 이득이 있다는 것을 확인하였다. 구체적으로 통합 메모리를 사용하는 임베디드 시스템 환경의 특성에 적합한 zero-copy기법을 적용하여 CPU와 GPU 모두 접근이 가능하도록 메모리 영역을 맵핑하는 방식으로 메모리 복제에 따른 오버헤드를 줄였으며, GoogLeNet 네트워크 모델에 대하여 10%의 처리 속도 향상과, 36% 소비 전력 감소를 확인하였다.

• Journal of Korean Institute of Industrial Engineers
• /
• v.20 no.2
• /
• pp.77-90
• /
• 1994

Curve-net interpolation surface is one of the most popular method in engineering design. Therefore it is supported with many commercial CAD/CAM system. However, construction algorithm of curve-net interpolation surfaces is rarely opened to the public because of its copy-right. In this paper we establish a construction algorithm of curve-net interpolation surface so called sweeping surface which especially concentrates on trajectory of cross-section curve. We also show the method which can construct sweeping surfaces as NURB or Bezier mathematical models. Surfaces having the form of standard mathematical models are very useful for the application of joining, trimming, blending etc. The proposed surface interpolation scheme consists of four steps; (1) preparation of guide curves and section curves, (2) remeshing guide curves and section curves, (3) blending section curves after deformation, and (4) determination of control points for sweeping surface using gordon method. The proposed method guarantee $G^1$-continuety, and construct the surface salifying given section curves and trajectory of section curves.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2002.11a
• /
• pp.306-310
• /
• 2002

Today, corporations are approach to inside information using public network such as Internet groping economical way to utilize public network to corporation information network without buying equipment. But, special quality of internet, Virtual Private Net (VPN) that it uses net that is observed as technology that can be guaranteed public safety division transmission and data securitybecause of can not secure data transmission. In this paper. add VPN function ONU&OLT of ATM-PON system and propose SCB (Single Copy Broadcasting). When there is VPN function to ATM network, the speed can be fast, and reduce rain track pick quantity during time more. Performance analysis network simulation that use NS-2.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.419-423
• /
• 2019

질문과 그에 대한 근거가 있는 문서를 읽고 정답을 예측하는 기계 독해 연구가 최근 활발하게 연구되고 있다. 기계 독해 문제를 위해 주로 사용되는 방법은 다층의 신경망으로 구성된 딥러닝 모델로 좋은 성능을 위해서는 양질의 대용량 학습 데이터가 필요하다. 그러나 질과 양을 동시에 만족하는 학습 데이터를 구축하는 작업에는 많은 경제적 비용이 소모된다. 이러한 문제를 해결하기 위해, 본 논문에서는 정답 분리 인코더와 복사 메커니즘을 이용한 단답 기반 한국어 질문 자동 생성 모델을 제안한다.

• BMB Reports
• /
• v.44 no.9
• /
• pp.578-583
• /
• 2011

Aspartate aminotransferase (AST) and alanine aminotransferase (ALT) are biochemical markers used to test for liver diseases. Copy number variation (CNV) plays an important role in determining complex traits and is an emerging area in the study various diseases. We performed a genome-wide association study with liver function biomarkers AST and ALT in 407 unrelated Koreans. We assayed the genome-wide variations on an Affymetrix Genome-Wide 6.0 array, and CNVs were analyzed using HelixTree. Using single linear regression, 32 and 42 CNVs showed significance for AST and ALT, respectively (P value < 0.05). We compared CNV-based genes between the current study (KARE2; AST-140, ALT-172) and KARE1 (AST-1885, ALT-773) using NetBox. Results showed 9 genes (CIDEB, DFFA, PSMA3, PSMC5, PSMC6, PSMD12, PSMF1, SDC4, and SIAH1) were overlapped for AST, but no overlapped genes were found for ALT. Functional gene annotation analysis shown the proteasome pathway, Wnt signaling pathway, programmed cell death, and protein binding.

• Journal of Advanced Navigation Technology
• /
• v.11 no.1
• /
• pp.112-117
• /
• 2007

The Internet leads the transformation of the all-over social life with its radical diffusion and development. Moreover, it can be more focussed on the electronic commerce using the Internet - a new type of commerce, which is diffusion and developing. In the paper, we propose an electronic payment protocol with a network-type electronic-cash based on Public Key Infrastructure(PKI). The proposed protocol overcomes the problem of NetBill which deals with only contents and can't ensure anonymity. It also prevents illegal copy and distribution and insures the greatest safety by means of giving a certification number to the digital contents offered on the on-line.

• KIISE Transactions on Computing Practices
• /
• v.22 no.2
• /
• pp.107-112
• /
• 2016

The Sejong Electronic(machine-readable) Dictionary, developed by the 21st century Sejong Plan, contains systematically organized information on Korean words. It helps to solve problems encountered in the electronic formatting of the still-commonly-used hard-copy dictionary. The Sejong Electronic Dictionary, however has a limitation relate to sentence structure and selection-restricted nouns. This paper discuses the limitations of word-sense disambiguation(WSD) that uses subcategorization information suggested by the Sejong Electronic Dictionary and generalized selection-restricted nouns from the Korean Lexico-semantic network. An alternative method that utilized semi-supervised learning, the chi-square test and some other means to make WSD decisions is presented herein.

• Journal of KIISE:Computing Practices and Letters
• /
• v.10 no.6
• /
• pp.484-498
• /
• 2004

In this paper. we propose a novel performance monitoring scheme for heterogeneous SOAP nodes. The scheme is basically based on two-level (kernel-level and user-level) packet filtering of TCP flows. By TCP flow, we mean a sequence of raw packet streams on a TCP transaction. In this scheme, we detect and extract SOAP operations embedded in SOAP messages from TCP flows. Therefore, it becomes possible to monitor heterogeneous SOAP nodes deployed on diverse SOAP-based middlewares such as .Net and Apache AXIS. We present two implementation mechanisms for the proposed scheme. The first mechanism tries to identify SOAP operations by analyzing all fragmented SOAP messages on TCP flows. However, a naive policy would incur untolerable overhead since it needs to copy all packets from kernel to user space. The second mechanism overcomes this problem by selectively copying packets from kernel to user space. For selective copying, we use a kernel-level packet filtering method that makes use of some representative TCP flags.(e.g. SIN, FIN and PSH). In this mechanism, we can detect SOAP operations only from the last fragment of SOAP messages in most cases. Finally, we implement a SOAP monitoring system using a component ca]led SOAP Sniffer that realizes our proposed scheme, and show experimental results. We strongly believe that our system will play a vital role as a tool for various services such as transaction monitoring and load balancing among heterogeneous SOAP nodes.