• 전자통신동향분석
• /
• 제33권4호
• /
• pp.1-13
• /
• 2018

Deep learning (DL) models have been widely applied to AI applications such image recognition and language translation with big data. Recently, DL models have becomes larger and more complicated, and have merged together. For the accelerated training of a large-scale deep learning model, model parallelism that partitions the model parameters for non-shared parallel access and updates across multiple machines was provided by a few distributed deep learning frameworks. Model parallelism as a training acceleration method, however, is not as commonly used as data parallelism owing to the difficulty of efficient model parallelism. This paper provides a comprehensive survey of the state of the art in model parallelism by comparing the implementation technologies in several deep learning frameworks that support model parallelism, and suggests a future research directions for improving model parallelism technology.

• 한국산업정보학회:학술대회논문집
• /
• 한국산업정보학회 1997년도 추계학술대회 발표논문집:21세기를 향한 정보통신 기술의 전망
• /
• pp.79-89
• /
• 1997

Logic programming has many advantages as a paradigm for parallel programming because it offers ease of programming while retaining high expressive power due to its declarative semantics. In parallel logic programming, one of the important issues is the compile-time parallelism detection. Static data-dependency analysis has been widely used to gather some information needed for the detection of AND-parallelism. However, the static data-dependency analysis cannot fully detect AND-parallelism because it does not provide some necessary functions such as the propagation of groundness. As an alternative approach, abstract interpretation provides a promising way to deal with AND-parallelism detection, while a full-blown abstract interpretation is not efficient in terms of computation since it inherently employs some complex operations not necessary for gathering the information on AND-parallelism. In this paper, we propose an abstract domain which can provide a precise and efficient way to use the abstract interpretation for the detection of AND-parallelism of logic programs.

• 한국생산제조학회지
• /
• 제23권6호
• /
• pp.642-646
• /
• 2014

This research describes the measurement of roll parallelism by a laser interferometer. Parallelism among rolls is an important factor for improving the precision of printing devices. A laser interferometer, which is a device for the precise measurement of distance, can be utilized to measure parallelisms between rolls. To measure distance between two rolls by using a laser interferometer, the laser must not be severed during measurement. To achieve this condition, a linear motion guide was installed to each roll being measured, and continuous measurement of distance between two rolls was implemented by the simultaneous control of two mirrors installed on the guides. The method to measure parallelism between two rolls presented in this research can be utilized to improve printing precision by enhancing parallelism between rolls in printing devices.

• 자연과학논문집
• /
• 제5권2호
• /
• pp.39-43
• /
• 1992

사용자의 관점에서 볼때는 compiler가 parallelism을 발견할 수 있게 하는것이 매우 요구되지만, 아무리 잘 만들어진 compiler 라 할지라도 conditional, functional 또는 I/O statement 등 프로그램내에 존재하는 많은 parallelism을 인식 하기가 결코 쉬운 일이 아니다. 경우에 따라서는 compiler의 parallelism 결정이 곤란할경우 사용자에게 feedback 시키는 경우가 종종 있다. 이런 입장에서 프로그래머는 parallelism을 바로 전달하기 위해서 프로그램의 일부를 재구성 시킬 필요가 자주 발생한다. 그러한 관점에서 Functional, Declarative 언어의 잇점이 있다고 할 수 있고, 그러기 위해서는 paralled machine에 적합한 Algorithm 이 필요하다. 그러나, 이미 사용중인 Algorithm이 Parallel Machine에 부적절 하다는 것을 의미 하는것은 아니다. 본 연구에서는, Fortran을 이용하여 Parallel Algorithm을 구현 시키기위한 Declarative 언어에 있어서 Array 및 Matrix 를 다루기위한 Abstraction 방법을 제시 하고자 한다.

• International Journal of Advanced Culture Technology
• /
• 제9권4호
• /
• pp.282-287
• /
• 2021

A multicore system must be able to take full advantage of the program's instruction and data parallelism. This study introduces the data replication technique as a support technique to maximize the program's instruction and data parallelism. Instruction level parallelism can be limited by data dependency. In this case, if data is replicated to each processor core and used, instruction level parallelism can be used to the maximum. The technique proposed in this study can maximize the performance improvement effect when applied to scientific applications such as matrix multiplication operation.

• 한국정보과학회논문지:시스템및이론
• /
• 제29권12호
• /
• pp.655-664
• /
• 2002

데이타 병렬 모델은 대규모 병렬성을 용이하게 얻을 수 있는 장점이 있지만, 데이타 분산으로 인한 통신 지연시간은 상당한 부담이 된다. 본 논문에서는 데이타 병렬 프로그램에 내재되어 있는 태스크 병렬성을 추출하여 이러한 통신 지연시간을 감추는데 이용할 수 있음을 보인다. 기존의 태스크 병렬성 추출은 데이타 병렬성을 고려하지 않았지만, 여기서는 데이타 병렬성을 그대로 유지하면서 태스크 병렬성을 활용하는 방법에 대해 설명한다. 데이타 병렬 루프를 포함할 수 있는 다수의 태스크 스레드들로 구성된 다중스레드 프로그램을 표현하기 위해 본 논문에서는 PCFG(Parallel Control Flow Graph)라는 표현 형태를 제안한다. PCFG는 단일 스레드인 원시 데이타 병렬 프로그램으로부터 HDG(Hierarchical Dependence Graph)를 통해 생성될 수 있으며, 또한 PCFG로부터 다중스레드 코드를 쉽게 생성할 수 있다.

• 한국정밀공학회:학술대회논문집
• /
• 한국정밀공학회 2005년도 춘계학술대회 논문집
• /
• pp.792-795
• /
• 2005

The guide-ways of precision machine tools are one of important element of machine tools. It has usually a pair of surfaces for constraint of one direction with bearing. In the case of precision machine tools, non-contact bearing such as hydrostatic bearing and aerostatic bearing is adopted usually. In this case, profiles of rails has effect on straightness and the clearance of bearing has effect on stiffness of guide way, which changes to higher if clearance changes to smaller. The clearance is varied along moving table according to relative distance of pair of rails. The relative distance of pair of rail can be divided by three properties. First and second properties are straightness of each pair of rail and bearing pad. And, third is parallelism about pair of rails and pairs of bearing pad. There are several methods for measuring straightness of each surface such as reversal method, sequential two point method, and way straightness. These straightness measuring methods are always acquiring deviation of profile from eliminating linear fitted inclined line and don't have the information of parallelism. Therefore, to get the small clearance for high stiffness, the straightness of rail and bearing pad and parallelism about pair of rails and pair of bearing pads are measured for correction such as regrinding, reassembling and lapping. In this research, new and easy method for measuring parallelism of pair of rails is suggested. Two displacement probe and sensor stage, which is carry on the displacement sensor, are needed. The simulation and experiment was accomplished about pair of horizontal guide way to confirm the measurement of parallelism. And, the third probe is added to measure the straightness of each rails by sequential two point method. From the estimation of combined these two methods, it is confirmed that the profiles of a pairs of rails can be measured.

• 한국정밀공학회지
• /
• 제24권3호
• /
• pp.117-123
• /
• 2007

This paper describes a three-probe system that can be used to measure the parallelism and straightness of a pair of rails simultaneously. The parallelism is measured using a modified reversal method, while the straightness is measured using a sequential two-point method. The measurement algorithms were analyzed numerically using a pair of functionally defined rails to validate the three-probe system. Tests were also performed on a pair of straightedge rails with a length of 250 mm and a maximum straightness deviation of $0.05{\mu}m$, as certified by the supplier. The experimental results demonstrated that the parallelism-measurement algorithm had a cancellation effect on the probe stage motion error. They also confirmed that the proposed system could measure the slope of a pair of rails about $0.06{\mu}rad$. Therefore, by combining this technique with a sequential differential method to measure the straightness of the rails simultaneously, the surface profiles could be determined accurately and eliminate the stage error. The measured straightness deviation of each straight edge was less than $0.05{\mu}m$, consistent with the certified value.

• 컴퓨터교육학회논문지
• /
• 제8권1호
• /
• pp.93-103
• /
• 2005

대부분의 프로그램 실행 시간은 루프 구조에서 소비되기 때문에 순차 루프 프로그램으로부터 병렬성을 추출하는 것은 프로그램을 빠르게 실행하는 데 필수적이다. 병렬성을 추출하기 위한 기존의 연구들은 주로 불변 자료 종속 거리에 초점을 맞추어왔다. 본 논문에서는 중첩 루프에서 자료 종속성을 제거하는 방법과 자료 종속성 제거 방법을 확장한 프로시저 호출을 가진 루프에서 병렬성을 추출하는 방법을 제안한다. 이 두 가지 방법들은 모두 자료 종속 거리에 관계없이 적용할 수 있다. 중첩 루프에서의 자료 종속성 제거 방법과 프로시저 호출을 가진 루프에서 병렬성을 추출하는 방법을 기존의 방법들과 CRAY-T3E에서 성능 평가를 하였다. 두 개의 방법 모두가 기존의 방법들보다 매우 우수함을 보였다.

• 한국정밀공학회:학술대회논문집
• /
• 한국정밀공학회 2003년도 춘계학술대회 논문집
• /
• pp.1051-1054
• /
• 2003

For micro-weighing, null balance method is widely used in the precision industrial fields, such as biomedicine, semiconductor, chemistry and so on. In order to obtain high resolution and large measurement range simultaneously, the mechanism should be analyzed and optimized. However, large measurement error can be generated according to the mass loading position and this error is called as a corner loading error. The corner loading error is caused by the parallelism error of a Roberval mechanism used to minimize it. The corner loading error is one of the most dominant error sources that should be removed. It is possible to design that the mechanism has no corner loading error theoretically, but the mechanism of the micro weighing device is very difficult to be realized as original design due to assembling and manufacturing error. For the required specification of the device, the precise manufacturing technique under a few $\mu\textrm{m}$ is required for the realization of the design. In this paper, the effects of the parallelism error are analyzed by using Lagrange method and verified by experiment. Also, the compensation mechanism is proposed and the corner loading error is reduced by restoring tile parallelism.