深度神经网络训练过程中：首先是进行初始化，根据需求设置神经网络的基本结构；然后进行前向传递（feedforward），层与层之间进行传递，求得误差；然后进行反向传播（back propogation），根据误差最小化原则，使用随机梯度下降法，对各个参数进行求导，确定下降方向，对各个参数进行更新(In the training process of deep neural network, firstly, initialization is carried out, and the basic structure of the neural network is set up according to the demand; secondly, feedforward is carried out, and errors are obtained by transferring between layers; then back propogation is carried out, and random gradient is used according to the principle of error minimization. The descent method is used to derive the parameters, determine the descent direction and update the parameters.)