Why sets Matlab automatically the activation functions for a neural network like this?

I am asking myself why chooses Matlab always automatically for the hidden layer tan-Sigmoid and for the output layer pureline as an activation function?

If it refers to a study, which discovers, that those activation functions are more efficient than the other, please let me know.

Answer 1

[Greg Heath]

That is a standard configuation for a neural net. It's operation is explained in every elementary text.

Thank you for formally accepting my answer

Greg

[Osama Tabbakh]

Could you please refer to one of this elementary text?

[Greg Heath]

Sorry, I lost all of my several hundred books via a moving van error..

See your library.

Greg

[Greg Heath]

MAYBE I SHOULD HAVE SAID any ELEMENTARY TEXT ON NNs.

GREG

Answer 2

[Greg Heath]

The simplest useful approximation is is a series of blocks with different heights and widths.

The simplest useful DIFFERENTIABLE approximation is is a series of ROUNDED blocks with different heights and lengths.

Combining sigmoids fits the bill!

GREG

[Osama Tabbakh]

Thank you for your answer, I do not need it just to understand it, but I need some references because I am writing right now my thesis.

Answer 3

[Sai Bhargav Avula]

As mentioned by others thats the default setup in MATLAB.

Coming to comparision between different activation functions.

It is generally recommended to use ReLU as the activation function. If your model suffers form dead neurons during training we should use leaky ReLu or Maxout function.

The Sigmoid and Tanh are generally not preferred as they suffer with vanishing Gradient Problem which causes a lots of problems to train,degrades the accuracy and performance of a deep Neural Network Model.

[Osama Tabbakh]

I can't use ReLU because I am not using a deep neural network. I need just a reason, why are set by Matlab like that?