The proceedings contain 3 papers. The topics discussed include: cost-efficient construction of performance models;using benchmarking and regression models for predicting CNN training time on a GPU;and benchmarking mac...
ISBN:
(纸本)9798400706455
The proceedings contain 3 papers. The topics discussed include: cost-efficient construction of performance models;using benchmarking and regression models for predicting CNN training time on a GPU;and benchmarking machine learning applications on heterogeneous architecture using reframe.
parallelism is now a standard from the hardware standpoint considering the multicore nature of current processors and their vector processing feature. Beside somehow following Moore's Law and/or running faster, th...
详细信息
ISBN:
(纸本)9798331506742;9798331506735
parallelism is now a standard from the hardware standpoint considering the multicore nature of current processors and their vector processing feature. Beside somehow following Moore's Law and/or running faster, the design of multicore processors was also driven by energy concern. With an increasing number of cores per chip and wider SIMD capabilities, the question of how the corresponding parallelism is related to power consumption is important for energy-aware parallel implementation. The goal of this work is to provide from experimentation some basic insights on the power consumption pattern related to the aforementioned levels of parallelism including accelerated-computing with GPUs.
Purpose This study aims to enhance the parallel performance of a parallel-in-space-and-time (PinST) finite-element method (FEM) using time step overlapping. The effectiveness of the developed method is clarified in a ...
详细信息
Purpose This study aims to enhance the parallel performance of a parallel-in-space-and-time (PinST) finite-element method (FEM) using time step overlapping. The effectiveness of the developed method is clarified in a magnet eddy-current loss analysis of a practical interior permanent magnet synchronous motor (IPMSM) using a massively parallelcomputing environment. Design/methodology/approach The developed PinST FEM is a combination of the domain decomposition method as a parallel-in-space (PinS) method and a parallel time-periodic explicit error correction (PTP-EEC) method, which is one of the parallel-in-time (PinT) approaches. The parallel performance of the PinST FEM is further improved by overlapping the time steps with different processes in the PTP-EEC method. Findings By applying the overlapping PTP-EEC method, the convergence of the transient solution to its steady state can be accelerated drastically. Consequently, the good parallel performance of the PinST FEM is achieved in magnetic field analyses of the practical IPMSM using a massively parallelcomputing environment, in which over 10,000 processes are used. Originality/value In this study, the PinST FEM based on time step overlapping is newly developed and its effectiveness is demonstrated in a massively parallelcomputing environment, in which using either the PinS or PinT method alone cannot achieve sufficient parallel performance. This finding implies a new direction of parallelcomputing approaches for electromagnetic field computation.
作者：
Butola, RajatLi, YimingKola, Sekhar ReddyNational Yang Ming Chiao Tung University
Parallel and Scientific Computing Laboratory Electrical Engineering and Computer Science International Graduate Program Hsinchu300093 Taiwan Institute of Pioneer Semiconductor Innovation
The Institute of Artificial Intelligence Innovation National Yang Ming Chiao Tung University Parallel and Scientific Computing Laboratory Electrical Engineering and Computer Science International Graduate Program The Institute of Communications Engineering the Institute of Biomedical Engineering Department of Electronics and Electrical Engineering Hsinchu300093 Taiwan
In this work, a dynamic weighting-artificial neural network (DW-ANN) methodology is presented for quick and automated compact model (CM) generation. It takes advantage of both TCAD simulations for high accuracy and SP...
详细信息
This tutorial deals with the integration of data engineering with network management and orchestration in telecommunication networks. It provides participants with a comprehensive insight into the use of data engineer...
详细信息
ISBN:
(纸本)9798400704130
This tutorial deals with the integration of data engineering with network management and orchestration in telecommunication networks. It provides participants with a comprehensive insight into the use of data engineering to improve the efficiency and performance of telecommunication systems, especially through the use of Artificial Intelligence (AI)/ Machine Learning (ML) technologies in network infrastructures. Practical applications are also demonstrated using relevant case studies to illustrate the implementation of these concepts.
The idea and implementation of a testbed are essen-tial steps in assessing and improving the functionality of wireless systems in real-time scenarios. In this research, we proposed a testbed for establishing a secure ...
详细信息
The proliferation of diverse network technologies has enabled concurrent utilization of multiple network media, such as WiFi and 4G/5G links on modern smartphones. While this advancement facilitates simultaneous downl...
详细信息
ISBN:
(纸本)9798331522735;9798331522728
The proliferation of diverse network technologies has enabled concurrent utilization of multiple network media, such as WiFi and 4G/5G links on modern smartphones. While this advancement facilitates simultaneous downloading of varied network contents, traditional multi-connection management methods, like reception-driven requesting schemes, often struggle to adapt to dynamic network conditions. This paper introduces a novel timer-driven requesting scheme applied to both parallel TCP and parallel MPTCP (Multipath TCP) methods, addressing the limitations of existing approaches. Unlike previous studies that relied solely on simulations, we present an experimental evaluation of this scheme on real machines. Our extensive experimentation reveals significant advantages of the timer-driven requesting scheme, particularly its superior adaptability in fluctuating network environments. These findings not only demonstrate the scheme's practical viability but also underscore its potential to revolutionize content retrieval in multi-network scenarios, paving the way for more efficient and resilient data transfer mechanisms in increasingly complex network ecosystems.
In order to address the conflict between operating cost and environmental impact, this paper proposes a parallel membrane computing-based and fuzzy multi-objective optimization approach for solving the dynamic economi...
详细信息
ISBN:
(纸本)9798350375794;9798350375800
In order to address the conflict between operating cost and environmental impact, this paper proposes a parallel membrane computing-based and fuzzy multi-objective optimization approach for solving the dynamic economic dispatch problem in a combined heat and power plant with wind power (CHPEED). By defining membership functions for the generation cost and pollutant gas emissions, the multi-objective problem is fuzzified and transformed into a single-objective nonlinear optimization problem using the maximum fuzzy satisfaction method. The parallel membrane computing strategy is applied to improve the traditional multi-objective optimization approach. To validate the effectiveness of this optimization method, the approach is applied to a 7-unit test system, and simulation results show that considering pollutant gas emissions in the dispatch model can achieve a balanced optimization of economic and emission reduction objectives compared to CHPED without considering environmental factors.
Carrier synchronization is traditionally implemented in analog method in parallel operation of power modules. Carrier connection of modules is via wires paralleled with communication wires of modules. These signals ma...
详细信息
Carrier synchronization is traditionally implemented in analog method in parallel operation of power modules. Carrier connection of modules is via wires paralleled with communication wires of modules. These signals may be mutual interfered and affected by power circuit. Therefore, reliability of parallel system is reduced. This paper developed a digital carrier synchronization method based on CAN bus, in which realizing digital communication per carrier period and carrier synchronization of parallel power electronics modules through only one pair of CAN bus. The experiment uses TMS320F2812 with CAN peripherals as control chip and designed two parallel inverters. The experiment verified that the proposed method is feasible and has good anti-interference ability.
When I started working on analog computing for neural network systems in the 1980s, the question everyone feared to be asked at the end of their presentation was "couldn't this be done on a DSP processor?&quo...
详细信息
暂无评论