• 한국통신학회논문지
• /
• 제25권4B호
• /
• pp.661-674
• /
• 2000

주기억장치 DBMS(MMDBMS)는 디스크가 아닌 주기억장치를 주요 저장 매체로서 사용하므로 고속의 처리를 요구하는 다양한 데이터베이스 응용을 효과적으로 지원한다. 본 논문에서는 차세대 MMDBMS Tachyon의 인덱스 관리자 개발에 관하여 논의한다. 인덱스 관리자는 객체에 대한 빠른 검색 기능을 지원하는 필수적인 DBMS 서브 컴포넌트이다. 기존의 연구 결과로서 다양한 인덱스 구조가 제안된 바 있으나, 실제 DBMS 상에서 인덱스 관리자를 개발하는 경우에 발생하는 실질적인 이슈들에 대해서는 거의 언급하고 있지 않다. 본 논문에서는 Tachyon의 인덱스 관리자의 개발 중에 경험한 실질적인 구현 이슈들을 언급하고, 이들에 대한 해결 방안을 제시한다. 본 논문에서 다루는 주요 이슈들은 (1) 인덱스 엔트리의 집약적 표헌, (2) 가변 길이 키의 지원, (3) 다중 애트리뷰트 키의 지원, (4) 중복 키의 지원, (5) 외부 API의 정의, (6) 동시성 제어, (7) 백업 및 회복 등이다. 본 연구 결과를 통하여 향후 MMDBMS 개발자들의 시행 착오를 최소화할 수 있으리라 생각된다.

• 정보과학회 논문지
• /
• 제44권1호
• /
• pp.51-56
• /
• 2017

질의응답 문제를 인공지능 모델을 통해 해결하는 연구는 메모리 네트워크의 등장으로 인해 방법론의 변화를 맞이하고 있으며, 그 중 동적 메모리 네트워크(DMN)는 인간 기억 체계에 착안하여 신경망 기반의 주의 기제를 적용하면서, 질의응답에서 일어나는 각 인지 과정들을 모듈화 했다는 특징들을 갖는다. 본 연구에서는 부족한 학습 데이터를 확장 시키고, DMN이 내포하고 있는 시간 인식의 한계를 개선해 정답률을 높이고자 한다. 실험 결과, 개선된 DMN은 1K-bAbI 문제의 테스트 데이터에서 89.21%의 정답률과, 95%를 질의응답 통과의 기준의 정답률으로 가정할 때 12개의 과제를 통과하는 성능을 보여 정확도 면에서 기존의 DMN에 비해 13.5%p 만큼 더 높고, 4개의 과제를 추가로 통과하는 성능 향상을 보여주었다. 또한 뒤이은 실험을 통해, 데이터 내에서 비슷한 의미 구조를 가지는 단어들은 벡터 공간상에서 강한 군집을 이룬다는 점과, 일화 기억 모듈 통과 횟수와 근거 사실 수의 성능에 큰 영향을 미치는 직접적인 연관성을 발견하였다.

• 한국군사과학기술학회지
• /
• 제3권2호
• /
• pp.42-52
• /
• 2000

A torque control method of a switched reluctance motor for the position control of a tank gun is presented. One of the widely used torque control scheme, torque sharing function method, is investigated and a new torque sharing function method is proposed that extends the definition region of the conventional TSF to both the positive and negative torque production regions. By using this definition, all kinds of the control inputs that consider switching on/off angle control as well as the current profiling can be described. A parametrized representation of the current profiles is proposed by using a series of B-spline functions, which reduces memory requirement and enables additional controllers. Optimal determination of the TSFs are also investigated for various control objectives. Moreover, the comparison study of each objective is presented. Since this method generalizes all of the possible control input, the current and torque profiles obtained from the optimization are the most suitable control input that satisfy the objectives.

• 대한전자공학회논문지
• /
• 제27권5호
• /
• pp.775-783
• /
• 1990

In this paper we implemented the fast thinning unit by modifying the pipeline architecture which was proposed by Stanley R. Sternberg. The unit is useful in preprocessing such as image representation and pattern recognition etc. This unit is composed of interface part, local memory part, address generation part, thinning processing part and control part. In thinning processing part, we shortened the thinning part which performed by means of look up table using window mapping table. Thus we improved the weakness of SAP, in which the number of delay pipeline and window pipeline are equal to image column size. Two independent memorys using tri-state buffer enable the two direction flow of address generated by address generation part. This unit avoids the complexity of architecture and has flexibility of image size by means of simple modification of logic bits.

• 산업공학
• /
• 제14권3호
• /
• pp.227-235
• /
• 2001

There is a rapidly growing demand for wireless telecommunication. However, the number of usable channel is very limited. Therefore, the problem of channel assignment becomes more and more important to use channels as efficiently as possible. The objective of this paper is to develop an evolution program (EP) to find an efficient dynamic channel assignment method for minimum interference among the channels within reasonable time. The series of specific channel number is used as a representation of chromosome. The only changed chromosomes by crossover and mutation are evaluated in each generation to save computation time and memory for the progress of improved EP. We can easily differentiate the fitness value of each chromosome using proposed evaluation function. We also control the weighting factor of the mutation rate and the used number of elitist chromosomes for the speed of convergence to the optimal solution.

• 대한전자공학회논문지
• /
• 제23권5호
• /
• pp.681-691
• /
• 1986

To represent the relations of the data dependency and resource conflict among micro-operations(MOP's) in the compaction process of microprograms, we propose a DDM (data dependent matrix) representation method instead of the DAG (conventional directed acyclic graph). Also, we propose a global compaction algorithm of microprograms to prevent a kind of block copying by cutting the trace at a junction block. The DDM method and compaction algoristhm have been applied to the Lah's example. The results shows that the proposed algorithm is more efficient than the conventional algorithms in reducing in reducing the total execution time and control memory space.

• 한국국방경영분석학회지
• /
• 제29권2호
• /
• pp.61-80
• /
• 2003

A fire sequencing problem is considered. Fire sequencing problem is a kind of scheduling problem that seeks to minimize the overall time span under a result of weapon­target allocation problem. The assigned weapons should impact a target simultaneously and a weapon cannot transfer the firing against another target before all planned rounds are consumed. The computational complexity of the fire sequencing problem is strongly NP­complete even if the number of weapons is two, so it is difficult to get the optimal solution in a reasonable time by the mathematical programming approach. Therefore, a genetic algorithm is adopted as a solution method, in which the representation of the solution, crossover and mutation strategies are applied on a specific condition. Computational results using randomly generated data are presented. We compared the solutions given by CPLEX and the genetic algorithm. Above $7(weapon){\times}15(target)$ size problems, CPLEX could not solve the problem even if we take enough time to solve the problem since the required memory size increases dramatically as the number of nodes expands. On the other hand, genetic algorithm approach solves all experimental problems very quickly and gives good solution quality.

• 한국정보처리학회논문지
• /
• 제6권5호
• /
• pp.1166-1176
• /
• 1999

The spationtemporal databases support a historical informations as well as spatial managements for various kinds of objects in the real world, and can be efficiently used in many applications such as geographic information system, urban plan system, car navigation system. However it is difficult to represent efficiently historical operations with conventional database query language for spatial objects. In terms of cost for query processing, it also degenerates performance of query processing because of syntactic limitations which is innate in conventional query representation. So in this paper, we introduce a new query language, entitled as STQL, which has been extended on the basis of the most popular relational database query language SQL. And we implement as well as evaluate a spationtemporal query processing system that get a query written by STQL and then process it in a main memory.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2000년도 제15차 학술회의논문집
• /
• pp.262-262
• /
• 2000

This paper presents a new torque sharing function method that extends the definition region of the conventional TSF to both the positive and negative torque production regions. By using this definition, all kinds of the control inputs that consider switching on/off angle control as well as the current profiling can be described. A parametrized representation of the current profiles is proposed by using a series of B-spline functions, which reduces memory requirement and enables additional controllers. Optimal determination of the TSFs are also investigated for various control objectives. Moreover, the comparison study of each objective is presented. Since this method generalizes all of the possible control input, the current and torque profiles obtained from the optimization are the most suitable control input that satisfy the objectives.

• 한국통신학회논문지
• /
• 제31권10C호
• /
• pp.942-947
• /
• 2006

In this paper, a new feature representation technique called Improved 2-directional 2-dimensional direct linear discriminant analysis (Improved $(2D)^2$ DLDA) is proposed. In the case of face recognition, thesmall sample size problem and need for many coefficients are often encountered. In order to solve these problems, the proposed method uses the direct LDA and 2-directional image scatter matrix. Moreover the selection method of feature vector and the method of similarity measure are proposed. The ORL face database is used to evaluate the performance of the proposed method. The experimental results show that the proposed method obtains better recognition rate and requires lesser memory than the direct LDA.