摘要
针对过程神经网络训练时间长的问题,提出了一种基于MPI与OpenMP混合编程的过程神经网络算法,该算法基于标准梯度下降过程神经网络的批处理模式,应用MPI与OpenMP编程技术,在局域网内实现多台计算机组成机群,并进行了不同数量级下的样本集的训练。试验证明,在适当选取并行粒度的情况下,基于MPI与OpenMP编程技术的数据并行算法可以加快训练速度。
Aiming at the problem of long training time of process neural networks,a training algorithm based on MPI and OpenMP is proposed.The proposed algorithm based on batch processing of process neural network by the standard gradient descent,uses MPI and OpenMP Programming technology.In the local area network,a cluster is composed of multiple computers.Meanwhile,the sample training is carried out under different magnitude levels,the test result shows that it can accelerate the training speed of process neural networks,when the parallel granularity is appropriate.
出处
《长江大学学报（自科版）（上旬）》
CAS
2010年第3期417-419,共3页
JOURNAL OF YANGTZE UNIVERSITY (NATURAL SCIENCE EDITION) SCI ＆ ENG
