说明：  对于许多问题，开始可能只需要一个隐藏层就能达到不错的结果，比如对于复杂的问题我们可以在隐藏层上使用足够多的神经元就可以了。很长一段时间人们满足于此，没有深入探索深度神经网络。但深度神经网络有更高的参数效率，神经元个数可以指数级减少，并且训练速度更快！
(For many problems, it may only take a hidden layer to achieve good results at first. For complex problems, we can use enough neurons on the hidden layer. For a long time people were content with this and did not explore deep neural networks in depth. But deep neural networks have higher parameter efficiency, the number of neurons can be reduced exponentially, and the training speed is faster!)