The denominator of generally neural network output often tends to be zero,leading to infinite loop when training sample data.The reliability of results is debased.

• 常规神经网络在当训练样本时分母项易趋于0,导致运算进入死循环,降低了结果的可信度。