• Title/Summary/Keyword: 조밀한 모델

Search Result 72, Processing Time 0.028 seconds

Implementation of FPGA-based Accelerator for GRU Inference with Structured Compression (구조적 압축을 통한 FPGA 기반 GRU 추론 가속기 설계)

  • Chae, Byeong-Cheol
    • Journal of the Korea Institute of Information and Communication Engineering
    • /
    • v.26 no.6
    • /
    • pp.850-858
    • /
    • 2022
  • To deploy Gate Recurrent Units (GRU) on resource-constrained embedded devices, this paper presents a reconfigurable FPGA-based GRU accelerator that enables structured compression. Firstly, a dense GRU model is significantly reduced in size by hybrid quantization and structured top-k pruning. Secondly, the energy consumption on external memory access is greatly reduced by the proposed reuse computing pattern. Finally, the accelerator can handle a structured sparse model that benefits from the algorithm-hardware co-design workflows. Moreover, inference tasks can be flexibly performed using all functional dimensions, sequence length, and number of layers. Implemented on the Intel DE1-SoC FPGA, the proposed accelerator achieves 45.01 GOPs in a structured sparse GRU network without batching. Compared to the implementation of CPU and GPU, low-cost FPGA accelerator achieves 57 and 30x improvements in latency, 300 and 23.44x improvements in energy efficiency, respectively. Thus, the proposed accelerator is utilized as an early study of real-time embedded applications, demonstrating the potential for further development in the future.

3D Model Retrieval Based on Ray Casting Technique (레이 캐스팅 기법을 기반으로 한 3차원 모델 검색)

  • Lee, Sun-Im;Hwang, Hye-Jung;Moon, Young-Shik
    • Proceedings of the Korean Information Science Society Conference
    • /
    • 2005.07b
    • /
    • pp.844-846
    • /
    • 2005
  • 본 논문에서는 3차원 모델을 검색하기 위한 형태 기반 기술자를 추출하는 알고리즘을 제안한다. 제안하는 알고리즘은 모델의 기하학적인 특성을 고려하여 레이 캐스팅 샘플링과 구면조화함수를 이용하는 방법이다. 레이 캐스팅 기법은 모델의 외형을 고려한 적응적인 방법으로 샘플링 하여 기술자에 포함되는 형태 정보를 증가시켜 기술자의 식별성을 높인다. 또한 구면조화함수 계수 추출에서는 기하학적인 주파수 특성을 고려하여 적응적인 계수를 추출한다. 이 방법은 검색 성능에 영향을 미치지 않고 기술자를 조밀하고 간결하게 만든다. 최종적으로 두 방법을 결합함으로써 검색 엔진에서 이용 가능한 기술자를 생성한다. 실험 결과에서는 제안된 알고리즘의 성능이 기존 방법에 비해 약 $12.5\%$ 향상 된 것을 확인 할 수 있다.

  • PDF

Numerical Investigation of Effects of Tip Clearance Height on Fan Performance and Tip Clearance Flow in an Axial Fan of the Cooling Tower (냉각탑용 축류팬의 팁 간격이 팬 성능 및 틈새 유동에 미치는 영향에 관한 수치해석적 연구)

  • Oh, Keon-Je
    • Journal of Power System Engineering
    • /
    • v.16 no.1
    • /
    • pp.44-50
    • /
    • 2012
  • 팁 간격의 크기가 냉각탑용 축류팬의 성능과 누설 유동에 미치는 영향을 조사하기 위해서 서로 다른 2가지 팁 간격을 가진 경우에 대해서 점성유동을 해석하였다. 케이싱 내에서 작동하는 축류팬 주위의 유동을 연속방정식, Navier-Stokes 방정식 등을 지배방정식으로 사용하여 수치해석 하였다. 난류유동에 나타나는 레이놀즈 응력은 ${\kappa}-{\epsilon}$ 난류모델을 사용하여 계산하였다. 전체적으로 H형 격자계를 사용하였으며, 팁 주위의 유동을 해석하기 위해서 팁 영역 주위에 부분적으로 조밀한 격자를 두었다. 팁 간격이 증가하면 누설 유동의 증가로 인한 유동 손실의 증가로 전압상승과 수력효율이 감소하였다. 팬 직경에 대한 팁 간격이 0.4%에서 1.0%로 증가하면 전압상승 값이 약 10% 정도 감소하였으며, 수력효율은 약 3% 정도 감소하였다. 팁 간격이 팁 근처 날개 주위의 압력에 미치는 영향을 보면, 팁 간격이 증가하여 누설 유동이 증가하면 흡입면과 압력면의 압력차가 전연 부근에서 감소함을 알 수 있었다. 누설 와류의 중심은 코드를 따라서 흡입면으로 부터 떨어져 나가면서 형성됨을 알 수 있었다. 누설 와류의 위치를 보면 팁 간격이 증가하면 와류 중심의 위치가 흡입면 쪽으로 이동하고, 흡입면에서 떨어진 거리도 날개 후반부에서 증가 폭이 커지는 포물선 형태로 증가함을 알 수 있었다.

Quantitative Precipitation Estimation using Overlapped Area in Radar Network (레이더의 중첩관측영역을 활용한 정량적 강수량 추정)

  • Choi, Jeongho;Han, Myoungsun;Yoo, Chulsang;Lee, Jiho
    • Journal of Wetlands Research
    • /
    • v.19 no.1
    • /
    • pp.112-121
    • /
    • 2017
  • This study proposed the quantitative precipitation estimation method using overlapped area in radar network. For this purpose, the dense rain gauges and radar network are used. As a result, we found a reflectivity bias between two radar located in different area and developed the new quantitative precipitation estimation method using the bias. Estimated radar rainfall from this method showed the apt radar rainfall estimate than the other results from conventional method at overall rainfall field.

Structural Dynamics Modification Using Surface Grooving Technique : Modified Algorithm and Result of Fine HDD Cover Model (개선된 알고리즘을 이용한 그루브를 통한 표면형상변형 동특성 변경법)

  • Park, Mi-You;Park, Young-Jin;Park, Youn-Sik
    • Proceedings of the Korean Society for Noise and Vibration Engineering Conference
    • /
    • 2005.11a
    • /
    • pp.834-837
    • /
    • 2005
  • Structural Dynamics Modification (SDM) is a very effective technique to improve structure's dynamic characteristics by adding or removing auxiliary structures, changing material properties and shape of structure. Among those of SDM technique, the method to change shape of structure has been mostly relied on engineer's experience and trial-and-error process which are very time consuming. In order to develop a systematic method to change structure shape, surface grooving technique is studied and successfully applied to HDD cover model. At first, to check the effect of mesh size, surface grooving technique was tested to the fine HDD cover FEmodel. And fur the more efficient method, the algorithm is modified. Removing the low-valued modal strain energy element among the target domain, computational effort can be greatly reduced and the result of simulation is similar with the other simulation result.

  • PDF

Inelastic Displacement Ratio for Strength-limited Bilinear SDF Systems (강도한계 이선형 단자유도 시스템의 비탄성 변위비)

  • Han, Sang-Whan;Lee, Tae-Sub;Seok, Seung-Wook
    • Journal of the Earthquake Engineering Society of Korea
    • /
    • v.14 no.4
    • /
    • pp.23-28
    • /
    • 2010
  • This study evaluated the effect of vibration, level of lateral yielding strength, site conditions, ductility factor, strain-hardening ratio, and post-capping ratio of the strength limited bilinear SDF systems on the inelastic displacement ratio. The nonlinear response history analysis was conducted using 240 ground motions which were collected at the sites classified as site classes B, C, and D according to the NEHRP. To account for the P-$\Delta$ effects, this study considered negative stiffness ratios ranging from -0.1 to -0.5 of elastic stiffness. Four different damping ratios are used: 2, 5, 10, and 20%. From this study, an equation of inelastic displacement ratio was proposed using nonlinear regression analysis.

Geometric Modeling and Data Simulation of an Airborne LIDAR System (항공라이다시스템의 기하모델링 및 데이터 시뮬레이션)

  • Kim, Seong-Joon;Min, Seong-Hong;Lee, Im-Pyeong;Choi, Kyung-Ah
    • Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
    • /
    • v.26 no.3
    • /
    • pp.311-320
    • /
    • 2008
  • A LIDAR can rapidly generate 3D points by densely sampling the surfaces of targets using laser pulses, which has been efficiently utilized to reconstruct 3D models of the targets automatically. Due to this advantage, LIDARs are increasingly applied to the fields of Defense and Security, for examples, being employed to intelligently guided missiles and manned/unmanned reconnaissance planes. For the prior verification of the LIDAR applicability, this study aims at generating simulated LIDAR data. Here, we derived the sensor equation by modelling the geometric relationships between the LIDAR sub-modules, such as GPS, IMU, LS and the systematic errors associated with them. Based on this equation, we developed a program to generate simulated data with the system parameters, the systematic errors, the flight trajectories and attitudes, and the reference terrain model given. This program had been applied to generating simulated LIDAR data for urban areas. By analyzing these simulated data, we verified the accuracy and usefulness of the simulation. The simulator developed in this study will provide economically various test data required for the development of application algorithms and contribute to the optimal establishment of the flight and system parameters.

A Study on the Effects of the Coefficient of Uniformity and Porosity on the Soil-Water Characteristic Curves of Sandy Soils (사질토의 함수특성곡선에 대한 균등계수와 공극율의 영향에 관한 연구)

  • Yoo, Kunsun
    • Journal of the Korean GEO-environmental Society
    • /
    • v.14 no.5
    • /
    • pp.41-47
    • /
    • 2013
  • Well-graded and dense soils have good engineering properties. Unsaturated soil properties such as shear strength, compressibility and permeability are closely related to the soil-water characteristic curve of the soil. Therefore it is important to study the effects of the coefficient of uniformity and the porosity on the soil-water characteristic curve of the sandy soils, which are also related to the grain size distribution and the density of the soil, respectively. In this study soil-water characteristic curves (SWCCs) for six sandy soil specimens were investigated using Tempe pressure cells. The test data were best-fitted to Fredlund and Xing equation. The obtained fitting parameters and the characteristic points of SWCCs were discussed and correlated with the porosity and the coefficient of uniformity of the specimens. The results show that the smaller the porosity of the specimen becomes, the larger the value of the residual matric suction becomes, whereas the larger the coefficient of uniformity of the specimen becomes, the larger the value of the residual matric suction becomes. Regardless of the coefficient of uniformity, the smaller the porosity of the specimen, the flatter the max. slope of SWCC.

TWO-Point Reactor Kinetics for Large D$_2$O Reflected Systems (다량의 중수반사체 계통에 대한 2-점노 운동방정식)

  • Noh, T.W.;Oh, S.K.;Kim, S.Y.;Kim, D.H.
    • Nuclear Engineering and Technology
    • /
    • v.19 no.3
    • /
    • pp.192-197
    • /
    • 1987
  • Two-point kinetic equations for a compact-core-with-bulky-D$_2$O-reflector system were developed. A unique feature of the system is that certain fission gammas create retarded photoneutrons in the D$_2$O reflector by (r, n) reaction. Coupling effect between the core and the reflector was investigated by simulating power transients with various ramp reactivity insertions. Special attention was paid to the phenomenon associated with spatial separation of photoneutrons and their precursors. Simulations show that accuracy of the two-point model is comparable with that of space-dependent approach. Also it is found that the explicily expressed photoneutron terms in the reflector equation slow down the power transient compared to non-photoneutron expressions. Detectors for reactor power control purpose prefer to be deployed in the core zone to be able to accurately perdict transient power.

  • PDF

3D Object Retrieval Based on Improved Ray Casting Technique (개선된 레이 캐스팅을 이용한 3차원 객체 검색 기법)

  • Lee Sun-Im;Kim Jae-Hyup;Moon Young-Shik
    • Journal of the Institute of Electronics Engineers of Korea CI
    • /
    • v.43 no.2 s.308
    • /
    • pp.72-80
    • /
    • 2006
  • In this paper, we propose a new descriptor for 3D model retrieval based on shape information. The proposed method consists of two steps including ray casting method and spherical harmonic function, considering geometric properties of model. In the ray casting method, an adaptive sampling is performed for external shape information. By increasing shape information included in the descriptor, we improve the discriminating power of the proposed descriptor. The coefficients of spherical harmonic function are adaptively calculated, considering geometric frequency characteristics. This makes the descriptor more compact and concise without decreasing the retrieval performance. By combining two methods, we achieve more improved retrieval results.