Different activation function to use with Convolutional Neural Network

Hi there,

I am trying to use "Step Function" as the activation function for one of the convolutional layers in the CNN, though the only option Matlab provides is ReLU Layer, how can I add a different activation function?

Also what if the output of the CNN is another image not a class! What should I use as the final layer: Classification Layer or Regression Layer?

[Karthiga Mahalingam]

You can create custom layers for neural network models in matlab. This link has examples for creating custom deep learning, classification and regression layers:

https://www.mathworks.com/help/nnet/ug/define-custom-deep-learning-layers.html

The Deep learning layer example explains how to construct a PReLU layer. You could modify it accordingly for a step function.

To use a classification or a regression layer as the final layer would depend on your application. The above link also has examples to creating custom classification and regression layers.