• 한국어정보학회:학술대회논문집
• /
• 2016.10a
• /
• pp.56-61
• /
• 2016

본 논문에서는 Sequence-to-sequence 모델을 생성요약의 방법으로 한국어 문서요약에 적용하였으며, copy mechanism과 input feeding을 적용한 RNN search 모델을 사용하여 시스템의 성능을 높였다. 인터넷 신문기사를 수집하여 구축한 한국어 문서요약 데이터 셋(train set 30291 문서, development set 3786 문서, test set 3705문서)으로 실험한 결과, input feeding과 copy mechanism을 포함한 모델이 형태소 기준으로 ROUGE-1 35.92, ROUGE-2 15.37, ROUGE-L 29.45로 가장 높은 성능을 보였다.

• Journal of KIISE
• /
• v.44 no.5
• /
• pp.503-509
• /
• 2017

In this paper, the copy mechanism and input feeding are applied to recurrent neural network(RNN)-search model in a Korean-document summarization in an end-to-end manner. In addition, the performances of the document summarizations are compared according to the model and the tokenization format; accordingly, the syllable-unit, morpheme-unit, and hybrid-unit tokenization formats are compared. For the experiments, Internet newspaper articles were collected to construct a Korean-document summary data set (train set: 30291 documents; development set: 3786 documents; test set: 3705 documents). When the format was tokenized as the morpheme-unit, the models with the input feeding and the copy mechanism showed the highest performances of ROUGE-1 35.92, ROUGE-2 15.37, and ROUGE-L 29.45.

• Annual Conference on Human and Language Technology
• /
• 2016.10a
• /
• pp.56-61
• /
• 2016

본 논문에서는 Sequence-to-sequence 모델을 생성요약의 방법으로 한국어 문서요약에 적용하였으며, copy mechanism과 input feeding을 적용한 RNN search 모델을 사용하여 시스템의 성능을 높였다. 인터넷 신문기사를 수집하여 구축한 한국어 문서요약 데이터 셋(train set 30291 문서, development set 3786 문서, test set 3705문서)으로 실험한 결과, input feeding과 copy mechanism을 포함한 모델이 형태소 기준으로 ROUGE-1 35.92, ROUGE-2 15.37, ROUGE-L 29.45로 가장 높은 성능을 보였다.

• Journal of KIISE:Computer Systems and Theory
• /
• v.32 no.11_12
• /
• pp.557-565
• /
• 2005

This paper presents the development and implementation of a zero-copy file transfer mechanism that improves the efficiency of file transfers for PC cluster systems using hardware-based VIA(Virtual Interface Architecture) network adapters. VIA is one of the representative user-level communication interfaces, but because there is no library for file transfer, one copy occurs between kernel buffer and user boilers. Our mechanism presents a file transfer primitive that does not require the file system to be modified and allows the NIC to transfer data from the kernel buffer to the remote node directly without copying. To do this, we have developed a hardware-based VIA network adapter, which supports the PCI 64bit/66MHz bus and Gigabit Ethernet, as a NIC, and implemented a zero-copy file transfer mechanism. The experimental results show that the overhead of data coy and context switching in the sender is greatly reduced and the CPU utilization of the sender is reduced to $30\%\~40\%$ of the VIA send/receive mechanism. We demonstrate the performance of the zero-copy file transfer mechanism experimentally. and compare the results with those from existing file transfer mechanisms.

• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.490-495
• /
• 2021

표준어와 방언사이에는 위계가 존재하지 않고 열등하지 않다는 사상을 기반으로 방언을 보존하기 위한 다양한 노력들이 이루어지고있다. 또한 동일한 국가내에서 표준어와 방언간의 의사소통이 잘 이루어져야한다. 본 논문은 방언 연구보존과 의사소통의 중요성을 바탕으로 한국어 방언 기계번역 연구를 진행하였다. 대표적인 방언 중 하나인 제주어와 더불어 강원어, 경상어, 전라어, 충청어 기반의 기계번역 연구를 진행하였다. 공개된 AI Hub 데이터를 바탕으로 Transformer기반 copy mechanism을 적용하여 방언 기계번역의 성능을 높이는 모델링 연구를 진행하였으며 모델배포의 효율성을 위하여 Many-to-one기반 universal한 방언 기계번역기를 개발하였고 이를 one-to-one 모델과의 성능비교를 진행하였다. 실험결과 copy mechanism이 방언 기계번역 모델에 매우 효과적인 요소임을 알 수 있었다.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.1 no.2
• /
• pp.131-136
• /
• 2008

From the reciprocal connection by this Internet network researchs about the efficiency improvement of the whole system is accomplished with the method which reduces delays in message transmission. From here, we will do a comparative study between the user data program protocol (UDP) and the zero copy which does not use the buffer cache to fine out the valid method to improve the efficiency. In this thesis, I will change the message copy from execution process of the buffer cache of the TCP/IP on Unix OS with process on Linux OS. The object of conversion is to show you that the zero copy which doesn't use the buffer cache from transfer control class improves the communication efficiency.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.28 no.2B
• /
• pp.129-137
• /
• 2003

Multimedia streams, like MPEG continuously retrieve multimedia data because of their incessant playback. While these streams need an efficient support of kernel, the current buffer cache mechanism of Linux kernel such as Unix operating system was designed apt for small files, which is aperiodically requested as well as time uncritical. But, in case of continuous media, the CPU must enormously copy memory from kernel address space to user address space. This must lead to a large CPU overhead. This overhead both degrades system throughput and cannot guarantee QOS. In this paper, we have designed and implemented two memory copy reduction schemes in Linux kernel, direct I/O and one copy. The direct I/O skips the buffer cache layer of Linux kernel and results in dramatic reduction of CPU memory copy overhead. And, the one copy provides a fast disk-to-network data path without copying to user address space. The experimental results show considerable reduction of CPU overhead and throughput improvements.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.301-306
• /
• 2019

문서 생성 요약은 최근 딥러닝을 이용한 end-to-end 시스템을 통해 유망한 결과들을 보여주고 있어 연구가 활발히 진행되고 있는 자연어 처리 분야 중 하나이다. 하지만 문서 생성 요약 모델을 구성하기 위해서는 대량의 본문과 요약문 쌍의 데이터 셋이 필요한데, 이를 구축하기가 쉽지 않다. 따라서 본 논문에서는 정교한 뉴스 기사 요약 데이터 셋을 기계적으로 구축하는 방법을 제안한다. 또한 딥러닝 기반의 생성 요약은 입력 문서와 다른 정보를 생성하거나, 또는 같은 단어를 반복하여 생성하는 문제점들이 존재한다. 이를 해결하기 위해 요약문을 생성할 때 입력 문서의 내용을 인용하는 복사-메커니즘과, 추론 단계에서 단어 반복을 직접적으로 제어하는 페널티를 사용하면 상대적으로 안정적인 문장이 생성될 수 있다. 그리고 Transformer 모델은 순환 신경망 모델보다 요약문 생성 과정에서 시퀀스 길이가 긴 본문의 정보를 적절히 인코딩하여 줄 수 있는 모델이다. 따라서 본 논문에서는 복사-메커니즘과 추론 단계의 페널티를 이용한 Copy-Transformer 모델을 한국어 문서 생성 요약 데이터에 적용하였다. 네이버 지식iN 질문 요약 데이터 셋과 뉴스 기사 요약 데이터 셋 상에서 실험한 결과, 제안한 모델을 이용한 생성 요약이 비교 모델들 대비 가장 좋은 성능을 보이고 양질의 요약을 생성하는 것을 확인하였다.

• Journal of Communications and Networks
• /
• v.9 no.1
• /
• pp.43-49
• /
• 2007

The implementation of client polling as a weak cache coherence mechanism has two major drawbacks: Firstly, the cache may return a stale copy if the object is changed in the origin server while the cached copy is considered valid. Secondly, the cache can invalidate a cached copy that is still valid in the server. Therefore, we propose a multilevel pre-fetching (MLP) in conjunction with the client polling to refine these drawbacks. MLP is introduced to improve the level of freshness among the cached objects. The simulation results presented in this paper show that the proposed MLP significantly minimizes the number of stale objects and reduces the invalidation messages sent out to the server, i.e., increase the cache HIT rate.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.25 no.5B
• /
• pp.899-912
• /
• 2000

PVM provides users with a single image of high performance parallel computing machine by collecting machines distributed over a network. Low communication overhead is essential to effectively run applications on PVM based platforms. In the original PVM, three times of memory copies are required for a PVM task to send a message to a remote task, which results in performance degradation. We propose a zero-copy model using global shared memory that can be accessed by PVM tasks, PVM daemon, and network interface card(NIC). In the scheme, a task packs data into global shared memory, and notify daemon that the data is ready to be sent, then daemon routes the data to a remote task to which it is sent with no virtual data copy overhead. Experimental result reveals that the message round trip time between two machines is reduced significantly in the proposed zero-copy scheme.