• Journal of the Korea Society of Computer and Information
• /
• v.11 no.3
• /
• pp.141-149
• /
• 2006

In this paper, we introduce an improved algorithm for computing matrix triple product that commonly arises in primal-dual optimization method. In computing $P=AHA^{t}$, we devise a single pass algorithm that exploits the block diagonal structure of the matrix H. This one-phase scheme requires fewer floating point operations and roughly half the memory of the generic two-phase algorithm, where the product is computed in two steps, computing first $Q=HA^{t}$ and then P=AQ. The one-phase scheme achieved speed-up of 2.04 on Intel Itanium II platform over the two-phase scheme. Based on memory latency and modeled cache miss rates, the performance improvement was evaluated through performance modeling. Our research has impact on performance tuning study of complex sparse matrix operations, while most of the previous work focused on performance tuning of basic operations.

• The Transactions of the Korea Information Processing Society
• /
• v.5 no.6
• /
• pp.1418-1430
• /
• 1998

Temporal databases support an efficient historical management by means of valid time and transaction time. Valid time stands for the time when a data happens in the real world. And transaction time stands for the time when a data is stored in the database, Temporal Query Processing System(TQPS) should be extended so as tc process the temporal operations for the historical informations in the user query as well as the conventional relational operations. In this paper, the extended temporal query processing systems which is based on the previous temporal query processing system for TQuel(Temporal Query Language) consists of the temporal syntax analyzer, temporal semantic analyzer, temporal code generator, and temporal interpreter is to be described, The algorithm for additional functions such as transaction time management, temporal aggregates, temporal views, temporal joins and the heuristic optimization functions and their example how to be processed is shown.

• Journal of the Korean Geographical Society
• /
• v.38 no.4
• /
• pp.575-590
• /
• 2003

It was the landscape phase that social relations were extended by production of landscapes during the 16C～17C. Core lineage groups in study area had extended socio-spatial nexus through making of the landscape of ‘authority-ostentation’(권력-과시형 경관) and of ‘kindness-civilization’(시혜ㆍ교화형 경관). Therefore they could gain a result that their power and authority took root in community perfectly. In the case of the Kims of Kwangsan(광산김씨), the first landscape of ‘kindness-civilization’ was Jungheodang(the learning room Jungheo), and next it was built Yangsungdang(the learning room Yangsung), and Imlijung(the pavilion Imli). Through this landscape production, the consciousness of scholastic genealogy was born, and this genealogy consciousness could be deepened and extended in regional scale. One more important feature in this phase is the point that the Kims of Kwangsan in Yeonsan area and the Songs of Unjin(은진송씨) in Hoiduk area could unite socially by the landscape production of ‘kindness-civilization type’ becoming intermediation. And these social union, that is to say, it do connote enlargement of the territoriality. It tells that community of ‘regional’ scale was formed by of core lineage groups of ‘local’ uniting each other.

• Proceedings of the IEEK Conference
• /
• 1998.06a
• /
• pp.665-668
• /
• 1998

본 논문에서는 초저속 동영상 부호화에 관한 것으로, 움직임 추정 효율을 개선하기 위해 분할된 영역별로 움직임 정보를 추정하여 부호화를 수행할 뿐만 아니라 분할된 영역중 의미있는 부분을 선택적으로 부호화할 수 있는 영역분할 기반 영상부호화 기법에 관한 것이다. 첫째로, 움직임 추정은 분할된 영역 정보를 이용한 가변 탐색 영역 설정을 통해 전역 탐색 움직임 추정시 소모되는 많은 연산량을 줄이고, 둘째로, 움직임 추정 후 추정된 움직임 정보를 이용해 영역의 재분할 과정을 통해 분할된 영역별로 움직임 정보를 부호화 함으로써 개선된 부호화 효율을 보이며, 셋째로, 분할된 영역 중 얼굴과 같은 의미를 갖는 영역을 선택적으로 부호화하고 전송하기 위한 인식 기법을 제안하고자 한다.

• Proceedings of the Computational Structural Engineering Institute Conference
• /
• 2009.04a
• /
• pp.146-150
• /
• 2009

구조 시스템 식별은 역문제로서 이상화된 유한요소 모델을 실험치와 일치시키기 위해 유한요소모델을 보정하는 형태로 주로 이루어진다. 이를 위해 비선형 섭동법이 사용되고 있으며 이 방법을 실제 문제에 사용하기 위해서 시스템 축소법에 대한 연구가 진행 되고 있다. 하지만 기존의 방법에서는 유한요소모델의 모든 요소가 실험치와 다르다고 가정하여서 전체 요소 수만큼의 설계 변수를 두어서 역해석을 수행한다. 이런 기존의 방법에서는 시스템이 커짐에 따라 연산 시간이 기하급수적으로 증가하게 되어 어려움이 있다. 설계 변수의 증가는 해공간(solution space)의 확장을 의미하며 이는 해의 정확성에 큰 영향을 끼친다. 본 연구에서는 모델을 적은 수의 설계영역으로 나누어서 반복연산 단계마다 해의 경향성을 이용해서 설계 영역을 전략적으로 변경하는 적응성 설계영역기법을 제안한다. 수치예제를 통해 본 연구에서 제안하는 기법의 정확도와 효용성을 고찰한다.

• The KIPS Transactions:PartA
• /
• v.9A no.3
• /
• pp.379-386
• /
• 2002

This paper presents a scheduling approach for synthesizing pipelined datapaths under resource constraints. Our approach builds up a schedule based on gradual mobility reduction in contrast to other algorithms of previous researches, where an operation being scheduled is selected by using a priority function. The proposed method consists of a scheduling algorithm and a decision algorithm for detecting any violation against resource constraints. Our approach evaluates whether or not a scheduling solution can exist in case an operation temporarily is assigned to the earliest or latest control step among the assignable steps for the operation. If a solution cannot be found, it is impossible to assign the operation to that step due to a violation against resource constraints, and so we can eliminate that control step. This process is iterated until a reduction of mobility for all operations can not be obtained. Experiments on benchmarks show that this approach gains a considerable improvement over those by previous approaches.

• Journal of KIISE:Databases
• /
• v.30 no.2
• /
• pp.145-156
• /
• 2003

Most of the content-based image retrieval systems focuses on similarity-based retrieval of natural picture images by utilizing color. shape, and texture features. For the neuroscience image databases, we found that retrieving similar images based on global average features is meaningless to pathological researchers. To realize the practical content-based retrieval on images in neuroscience databases, it is essential to represent internal contents or semantics of images in detail. In this paper, we present how to represent image contents and their related concepts to support more useful retrieval on such images. We also describe the operational semantics to support these advanced retrievals by using object-oriented message path expressions. Our schemes are flexible and extensible, enabling users to incrementally add more semantics on image contents for more enhanced content searching.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.16 no.1
• /
• pp.33-40
• /
• 2023

With the development of supercomputing technology and hardware technology, numerical computation methods are also being advanced. Accordingly, improved weather prediction becomes possible. In this paper, we propose to apply the LAPACK(Linear Algebra PACKage) BLAS(Basic Linear Algebra Subprograms) library to the linear algebraic numerical computation part within the source code to improve the performance of the cumulative parametric code, Unicon(A Unified Convection Scheme), which is included in SCAM(Single-Columns Atmospheric Model, simplified version of CESM(Community Earth System Model)) and performs standby operations. In order to analyze this, an overall execution structure diagram of SCAM was presented and a test was conducted in the relevant execution environment. Compared to the existing source code, the SCOPY function achieved 0.4053% performance improvement, the DSCAL function 0.7812%, and the DDOT function 0.0469%, and all of them showed a 0.8537% performance improvement. This means that the LAPACK BLAS application method, a library for high-density linear algebra operations proposed in this paper, can improve performance without additional hardware intervention in the same CPU environment.

• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.439-443
• /
• 2021

기계 독해는 단락과 질의가 주어졌을 때 단락 내 정답을 찾는 자연어 처리 태스크이다. 최근 벤치마킹 데이터셋에서 사전학습 언어모델을 기반으로 빠른 발전을 보이며 특정 데이터셋에서 인간의 성능을 뛰어넘는 성과를 거두고 있다. 그러나 이는 단락 내 범위(span)에서 추출된 정보에 관한 것으로, 실제 연산을 요구하는 질의에 대한 응답에는 한계가 있다. 본 논문에서는 기존 범위 내에서 응답이 가능할 뿐만이 아니라, 연산에 관한 이산 추론을 요구하는 단락 및 질의에 대해서도 응답이 가능한 기계 독해 모델의 효과성을 검증하고자 한다. 이를 위해 영어 DROP (Discrete Reasoning Over the content of Paragraphs, DROP) 데이터셋으로부터 1,794개의 질의응답 쌍을 Google Translator API v2를 사용하여 한국어로 번역 및 정제하여 KoDROP (Korean DROP, KoDROP) 데이터셋을 구축하였다. 단락 및 질의를 참조하여 연산을 수행하기 위한 의미 태그를 한국어 KoBERT 및 KoELECTRA에 접목하여, 숫자 인식이 가능한 KoNABERT, KoNAELECTRA 모델을 생성하였다. 실험 결과, KoDROP 데이터셋은 기존 기계 독해 데이터셋과 비교하여 단락에 대한 더욱 포괄적인 이해와 연산 정보를 요구하였으며, 가장 높은 성능을 기록한 KoNAELECTRA는 KoBERT과 비교하여 F1, EM에서 모두 19.20의 월등한 성능 향상을 보였다.

• Education of Primary School Mathematics
• /
• v.22 no.3
• /
• pp.181-203
• /
• 2019

Even though the properties of operations in multiplication serve a fundamental basis of conceptual understanding the multiplication with whole numbers for elementary students, there has been lack of research in this field. Given this, the purpose of this study was to analyze instructional methods related to the properties of operations in multiplication (i.e., commutative property of multiplication, associative property of multiplication, distributive property of multiplication over addition) in a series of mathematics textbooks of Korea, Japan, and the US. The overall analysis was conducted in the following two aspects: (a) when and how to deal with the properties of multiplication in three instructional context (i.e., introduction, application, generalization), and (b) what models use to represent the properties of multiplication. The results of this showed that overall similarities in introducing the properties of multiplication .in (one digit) ${\times}$ (one digit) as well as emphasizing the divers representation. However, subtle but meaningful differences were analyzed in applying and generalizing the properties of multiplication. Based on these results, this paper closes with some implications on how to teach the properties of operations in multiplication properties in elementary mathematics.