BP神经网络原理研究与实现

The Research and Implementation of the Principle of BP Neural Network

本文介绍了BP神经网络(BPNN)的结构模型及对神经网络进行训练和测试的有关概念,对BPNN向前传播及向后传播的数学原理进行了详细推导,对梯度下降算法进行了深入讨论。作为对BPNN的运用和检验,作者按照BPNN的原理编写了一款软件工具(BPNNsoft),并在MNIST数据集上进行训练和测试,论文最后对训练和测试情况进行了简要说明和分析。

In this paper, the structure model of BP neural network （BPNN） and the related concepts of training and testing of neural network are introduced. The mathematical principles of forward and backward propagation of BPNN are derived in detail, and the Gradient Descent Algorithm is discussed in depth. As the use and test of BPNN, the author wrote a software tool （BPNNsoft） in accordance with the principle of BPNN, and trained and tested on the MNIST data set. Finally, the paper gave a brief description and analysis of the training and test situation.