• 한국어정보학회:학술대회논문집
• /
• 2017.10a
• /
• pp.92-96
• /
• 2017

딥 러닝 모델은 여러 히든 레이어로 구성되며, 히든 레이어의 깊이가 깊어질수록 레이어의 벡터는 높은 수준으로 추상화된다. 본 논문에서는 Encoder RNN의 레이어를 여러 층 쌓은 멀티 레이어 포인터 네트워크를 제안하고, 멀티 태스크 학습 기반인 멀티 레이어 포인터 네트워크를 이용한 한국어 의존 구문 분석 모델을 제안한다. 멀티 태스크 학습 모델은 어절 간의 의존 관계와 의존 레이블 정보를 동시에 구하여 의존 구문 분석을 수행한다. 실험 결과, 본 논문에서 제안한 모델이 기존 한국어 의존 구문 분석 연구들 보다 좋은 UAS 92.16%, LAS 89.88%의 성능을 보였다.

• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.13-18
• /
• 2018

구문 분석은 문장의 구조를 이해하며 의미의 중의성을 해결하는 것이다. 일반적으로 한국어는 어순 배열의 자유도가 높고 문장 성분의 생략이 빈번한 특성이 있기 때문에 의존 구문 분석이 주된 연구 대상이 되어 왔다. 스택-포인터 네트워크 모델은 의존 구문 파서에 맞게 포인터 네트워크 모델을 확장한 것이다. 스택-포인터 네트워크는 각 단어에서 의존소를 찾는 하향식 방식의 모델로 기존 모델의 장점을 유지하면서 각 단계에서 파생된 트리 정보도 사용한다. 본 연구에서는 스택-포인터 네트워크 모델을 한국어에 적용해보고 이와 함께 어절 정보를 반영하는 방법을 제안한다. 모델의 실험 결과는 세종 구문 구조를 중심어 후위(head-final)를 엄격히 준수하여 의존 구문 구조로 변환한 것을 기준으로 UAS 92.65%의 정확도를 얻었다.

• Proceedings of the IEEK Conference
• /
• 2001.06a
• /
• pp.269-272
• /
• 2001

Since Multi-way Search Tree reduces the number of the memory access by increasing the branch factor, it is considered a method to archive the high-speed IP address lookup. Using the combination of initial 16 bit may and Multi-way Search Tree, it also reduces the search time of If address lookup. Multi-way Search Tree consists of K keys and K＋1 key pointers. This paper shows how the E update of Multi-way Search Tree which consists of the one pointer within a node can be performed. Using the one pointer within a node, it increases the number of keys within a node and reduces the search time of IP lookup. We also describes IP updating methods such as modification, Insertion and Deletion of address entries. Our update scheme performs better than the method which rebuilds the entire IP routing table when IP update processes.

• Proceedings of the KIEE Conference
• /
• 2007.04a
• /
• pp.373-376
• /
• 2007

As a part of automatic car parking, this paper have devised the distance measurement algorism using stereo vision system which measure between object and parking lot. In this simulation, laser pointer is used for detecting edge of designated object and real distance is measured by searching the position of laser pointer and measuring the distance. Which are 1 / 1.5 / 2 / 2.5 / 3m and adjusted correction had applied in order for use in car parking system.

• Journal of Advanced Navigation Technology
• /
• v.15 no.2
• /
• pp.227-233
• /
• 2011

This paper presents a small-area ISDB-T time deinterleaver structure. ISDB-T is an mobile TV standard that is widely used in Japan and many South American countries. One of the strong points of the standard is the long interleaving depth, which enhance the communication performance. However, long interleaving requires many delay buffers, in other words many pointer registers. This paper reduces the number of pointer registers with the deinterleaver equivalent transformation. The experimental results show that the area is reduced with the proposed structure.

• Journal of the Korean Society for Precision Engineering
• /
• v.16 no.6
• /
• pp.176-182
• /
• 1999

In this paper, we consider a decoupling control and a servo control for the JINAMCHA of which the cart and the pointer is coupled very strongly. The right eigenstructure assignment(i.e. simultaneous assignment of eigenvalues and right eigenvectors) is used for decoupling the motions of the cart and the pointer. The CGT(Command Generator Tracker) is used for a servo control of the JINAMCHA. The performance of the proposed control scheme, that is the combination of the right eigenstructure assignment and the CGT, is evaluated by simulations. The results show that the proposed controller has better performance than the LQR with limiting properties and the conventional LQ servo.

• Journal of Korea Society of Industrial Information Systems
• /
• v.20 no.5
• /
• pp.13-23
• /
• 2015

This thesis proposes a method for effectively detecting memory errors with low occurrence frequency that may occur depending on runtime situation by analyzing assembly codes obtained by disassembling an executable file. When applying the proposed method to various programs having no compilation error, a total of about 750 potential errors taken about 90 seconds are detected among 1 million lines of assembly codes corresponding to a total of about 10 thousand functions.

• Proceedings of the Safety Management and Science Conference
• /
• 2004.05a
• /
• pp.1-4
• /
• 2004

Need safety diagnostic indicator pointer that is objective as estimation basis to be scientific and manages systematically safety existing condition and level for city gas facilities. Safety diagnostic indicator pointer can divide by greatly two. One is management (Software) side that one that differ with mechanical (Hardware) side that handle laying stress on equipment connected with safety manages organization and management connected with safety. Weight appropriates according to specific gravity that relevant element occupies in facilities. To give so big value in particular element in grant of weight pays attention because break balance of whole guideline and should gives. Finally, efficient safety net may see effect before self-regulation inside by minimum target establishment by government and legal, systematic countermeasure establishment for this, strict law application and each business proprietor and though activity and user's active safety activity forms harmony properly.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.33 no.2
• /
• pp.319-323
• /
• 2023

In intra-process isolation, file descriptors work as another attack vector from the memory corruption attacks. The attacker can read or write by corrupting file descriptors so they can escape the isolation. In this paper, we propose new lightweight capability-based access control system on file descriptor using ARM's hardware extension, PA(Pointer Authentication). Our system was implemented on Linux kernel module, only shows 5% overhead to control the access on the file descriptor.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.543-545
• /
• 2019

본 논문은 최신 한국어 의존 구문 분석 모델(Korean dependency parsing model)들과 다양한 앙상블 모델(ensemble model)들을 결합하여 그 성능을 분석한다. 단어 표현은 미리 학습된 워드 임베딩 모델(word embedding model)과 ELMo(Embedding from Language Model), Bert(Bidirectional Encoder Representations from Transformer) 그리고 다양한 추가 자질들을 사용한다. 또한 사용된 의존 구문 분석 모델로는 Stack Pointer Network Model, Deep Biaffine Attention Parser와 Left to Right Pointer Parser를 이용한다. 최종적으로 각 모델의 분석 결과를 앙상블 모델인 Bagging 기법과 XGBoost(Extreme Gradient Boosting) 이용하여 최적의 모델을 제안한다.