Creating custom Transfer Function with more learnable parameters

Hello!

I actually working with deep learning toolbox.

I'm making a comparative study of the transfer functions and some of them aren´t in the toolbox, this functions that I would like to try have more learnable parameters than the weights and biases in a "classical" function. For example:

xtansig= 2./(1+exp(x*(-2*n))+1;

What functions will I have to modify to create a new transfer function that, in addition to weights and biases, also finds the "x"?

Answer 1

[Sam Chak]

Edit: This article contains information about creating custom functions for NN.

https://www.mathworks.com/help/deeplearning/ug/custom-neural-network-helper-functions.html

I think there are some similarities between your function and the logistic function. Have you looked into the logsig function?

https://www.mathworks.com/help/deeplearning/ref/logsig.html

[Alexia Irais]

Hello Sam.

Thank you for your answer.

I have searched in all the transfer functions that matlab has, I don't understand how I could generate the new function so that it finds more parameters :(