• Proceedings of the IEEK Conference
• /
• 2003.07b
• /
• pp.1137-1140
• /
• 2003

The area of the prototype device is less than 80mm$^2$. Operating with a 60ns clock cycle, the device typically dissipates only 300mW. The full functionality was proven by using the methodical test programs based on typical image processing operations. Also, we realized the whole process from conventional gray image to color image. Format converters, implemented using multidimensional access memories, transfer the data between the processing element array and conventional bit-parallel components in real time. The completed system is fully functional and performs typical low-level image processing tasks at speed exceeding 30 frames of traditional TV system per second.

• Journal of Institute of Control, Robotics and Systems
• /
• v.17 no.4
• /
• pp.328-335
• /
• 2011

GPU (Graphics Processing Units) is consists of SIMD (Single Instruction Multiple Data) architecture and provides fast parallel processing. A GA (Genetic Algorithm), which requires large computations, is implemented in GPU using CUDA (Compute Unified Device Architecture). Three kinds of execution models are presented according to different combinations of processing modules in GPU. Comparison experiments between GPU models and CPU are tested for a couple of benchmark problems by variation of population sizes and complexity of problem sizes.

• Journal of IKEEE
• /
• v.14 no.2
• /
• pp.69-75
• /
• 2010

A 3D graphics pipeline is largely divided into geometry stage and rendering stage. In this paper, we propose a method that accelerates a geometry processing in multi-core GP-GPU, using dual-phase structure. It can be improved by parallel data processing using SIMD of GP-GPU, dual-phase structure and memory prefetch. The proposed architecture improves approximately 19% of performance when it use all the features.

• Annual Conference of KIPS
• /
• 2018.05a
• /
• pp.44-47
• /
• 2018

시계열 데이터 처리를 위해 방대한 양의 데이터를 스토리지에서 빠르게 읽어와 처리하려는 움직임이 많아지고 있다. 이를 위해 스토리지의 read latency 를 개선하기 위한 여러 기법들이 제안되었지만, 이 기법들은 분산 노드의 스토리지 자원을 충분히 활용하지 못한다는 한계가 있다. 따라서 우리는 시계열 데이터를 실시간으로 처리하기 위해 스토리지에 병렬적으로 접근하여 read latency 를 개선하는 기법을 제안한다. 제안된 기법은 분산 환경에서 스토리지에 병렬적으로 접근하여, 각 노드에서 부분적으로 데이터를 읽어와 전체 데이터를 읽어오는 지연시간을 줄인다. 우리는 제안된 기법을 여러 노드로 구성된 분산 환경에서 구현하였다. 제안된 기법을 적용한 결과, 전체 데이터를 읽어오는 read latency 가 기존 기법보다 28.04% 줄어든 것을 확인하였다.

• Annual Conference of KIPS
• /
• 2012.11a
• /
• pp.207-208
• /
• 2012

SCADA 시스템과 같이 대규모의 데이터를 일정 시간이내에 처리하는 시스템 환경에서 가장 중요한 요소 중 하나가 성능이다. 사용자에게 직관적이며 편리한 UI를 제공하며 개발자는 유지보수성, 재사용성 등을 충분히 고려하여 시스템을 구현하여도 일정 성능 이상을 만족시키지 못한다면 사용할 수가 없다. 이러한 점을 고려하여 본 논문에서는 앞으로 SCADA 시스템이 감시, 제어하는 설비의 증가와 시스템 규모의 다양성 및 확장성을 갖춰야 함을 인식하고 다양한 성능향상 방법 중 소프트웨어 측면에서 병렬화 기법을 이용한 데이터 처리 방법을 소개한다.

• Journal of the Semiconductor & Display Technology
• /
• v.21 no.1
• /
• pp.75-78
• /
• 2022

Deep Neural Networks (DNN) has become an essential data processing architecture for the implementation of multiple computer vision tasks. Recently, DNN-based algorithms achieve much higher recognition accuracy than traditional algorithms based on shallow learning. However, training and inference DNNs require huge computational capabilities than daily usage purposes of computers. Moreover, with increased size and depth of DNNs, CPUs may be unsatisfactory since they use serial processing by default. GPUs are the solution that come up with greater speed compared to CPUs because of their Parallel Processing/Computation nature. In this paper, we analyze the inference time complexity of DNNs using well-known computer vision library, OpenCV. We measure and analyze inference time complexity for three cases, CPU, GPU-Float32, and GPU-Float16.

• The Journal of the Acoustical Society of Korea
• /
• v.39 no.6
• /
• pp.505-514
• /
• 2020

Since a large amount of training data is typically needed to train Deep Neural Networks (DNNs), a parallel training approach is required to train the DNNs. The Stochastic Gradient Descent (SGD) algorithm is one of the most widely used methods to train the DNNs. However, since the SGD is an inherently sequential process, it requires some sort of approximation schemes to parallelize the SGD algorithm. In this paper, we review various efforts on parallelizing the SGD algorithm, and analyze the computational overhead, communication overhead, and the effects of the approximations.

• Korean Journal of Air-Conditioning and Refrigeration Engineering
• /
• v.18 no.9
• /
• pp.698-706
• /
• 2006

Experiment and numerical calculation have been peformed to investigate mixed convection heat transfer between inclined parallel plates. Particle image velocimetry (PIV) with thermo-sensitive liquid crystal (TLC) tracers is used for visualizing and analysis. This method allows simultaneous measurement of velocity and temperature fields at a given instant of time. Quantitative data of the temperature and velocity are obtained by applying the color-image processing to a visualized image, and neural network is applied to the color-to-temperature calibration. The governing equations are discretized using the finite volume method. The results are presented for the Reynolds number ranges from 0.004 to 0.062, the angle of inclination, ${\Theta}$, from 0 to 45 degree and Prandtl number of the high viscosity fluid is 909. The results show velocity, temperature and mean Nusselt numbers distributions. It is found that the periodic flow of mixed convection between inclined parallel plates is shown at $0^{\circ}{\leq}{\Theta}<30^{\circ}$, Re<0.062, and the flow pattern can be classified into three patterns which depend on Reynolds number and the angle of inclination. The minimum Nusselt numbers occur at Re=0.05 regardless of the angle of inclination.

• KIPS Transactions on Software and Data Engineering
• /
• v.7 no.2
• /
• pp.63-68
• /
• 2018

Interests in face recognition area have been increasing due to diverse emerging applications. Face recognition algorithm from a two-dimensional source could be challenging in dealing with some circumstances such as face orientation, illuminance degree, face details such as with/without glasses and various expressions, like, smiling or crying. Hopfield Network capabilities have been used specially within the areas of recalling patterns, generalizations, familiarity recognitions and error corrections. Based on those abilities, a specific experimentation is conducted in this paper to apply the Redundant Parallel Hopfield Network on a face recognition problem. This new design has been experimentally confirmed and tested to be robust in any kind of practical situations.

• Journal of KIISE:Computing Practices and Letters
• /
• v.15 no.8
• /
• pp.586-590
• /
• 2009

The objective of this paper describes the design of MapReduce over Peer-to-Peer network for dynamic environments applications. MapReduce is a software framework used for Cloud Computing which processing large data sets in a highly-parallel way. Based on the Peer-to-Peer network character which node failures will happen anytime, we focus on using a DHT routing protocol which named Pastry to handle the problem of node failures. Our results are very promising and indicate that the framework could have a wide application in P2P network systems while maintaining good computational efficiency and scalability. We believe that, P2P networks and parallel computing emerge as very hot research and development topics in industry and academia for many years to come.