新型样条权函数神经网络的云计算研究

Research on Cloud Computing for Neural Network of a New Kind of Spline Weight Functions

采用权函数训练神经网络是近些年发展起来的一种算法。该算法有许多优点,例如可以直接求得全局最优点,有很好的泛化能力,训练后的权函数能够反映隐含在样本内部的有价值的信息特征等。因此进一步提高算法效率就显得十分重要。为了进一步提高运算速度,文中将神经网络与云计算相结合,采用云计算服务对一种新型的三次样条权函数神经网络算法的性能进行了分析,提出了云计算的定义,研究了三次样条权函数神经网络算法的并行机制。结果表明,采用云计算能够大幅提高三次样条权函数神经网络算法的效率。

Training neural network using weight functions is a new kind of algorithm developed in recent years, which has many advantanges, such as finding globe minima direcdy, good performance of generalization, extracting some useful information inherent in the problems and so on. It is very important to improve the efficiency for this new kind of algorithm. To improve the training speed, combine the neural networks with cloud computing, and analyze the performance of cloud computing services for the algorithm of training neural networks by cubic spline weight functions （NNCSWFs）. Attempt to give the definition for cloud computation, and study the parallel mechanism to improve the computing performance. The results indicate that, by using cloud computing services, the work efficiency for NNCSWFs algorithm is much higher than before.