Custom activation functions for Deep Learning Toolbox?

Hello,

I’m writing to ask whether it is possible to use custom activation functions with the deep learning toolbox, and the best way of using them if it is. It seems like this would be a good use for the custom layers feature, but I’ve been unable to determine whether it’s possible to use something other than image data with networks constructed in this way.

Answer 1

[Sebastian Castro]

Sure you can! Here is an example:

https://www.mathworks.com/help/deeplearning/ug/define-custom-deep-learning-layer.html

- Sebastian

[Andy]

does matlab support GELU — Gaussian Error Linear Unites ? (as described under: https://arxiv.org/pdf/1606.08415.pdf)