Why stack multiple hidden layers

A conclusion

When we need a better training model, we need smaller H (model parameters) and N (training data)

Why multiple hidden layers

Each of our function graphs can be represented by multiple activation functions + constants

When we want to represent a model, we can use a model with multiple hidden layers, or we can use a model with one layer, but the parameters generated by the model with multiple hidden layers willModels with far less than one layer

in order to getFor better results, when it is difficult for us to increase the data, try the method of multiple hidden layers to get better training results