What are the Learnable Parameters when Using Adaptive Activation


I would like to use the Adaptive activation function in my code as I had read some papers that this can help with convergence. I was able to update this in the configuration file but from what I am reading, besides enabling adaptive activation with a boolean value there should also be a hyper parameter I can change. Is there any documentation that I can reference to adjust the hyper parameters?

Thank you for your help,

Hi @tstone

There is some theory information on adaptive activations in our docs. The adaptive activation that is built into some of Modulus Symbolic models is a scalar one. As you mentioned right now there is only toggle control to turn this feature on and off (E.g. fully connected, note the learnable parameter here). Thus any customization or more advance scheme would require modification of the source code, but this should be fairly straight forward for most models.

Great, thank you for the information.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.