• Advertisement

Archived

This topic is now archived and is closed to further replies.

The "sharpness" of a sigmoid activation function

This topic is 5005 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Recommended Posts

Quick question: wouldn''t it be useful if the p in the sigmoid function 1/(1+e^(-a/p)), used as the activation function in a neural net, could be "evolved" along with the weights? It WILL affect the output, so wouldn''t it make sense to change it too (like in those experiments where they''re trying to evolve the topology of the nets)?

Share this post


Link to post
Share on other sites
Advertisement
Guest Anonymous Poster
Nope. It doesn''t really make much sense to change its value. For example, if you doubled it, it would have the same effect as halving all the weights, and vice versa.

Share this post


Link to post
Share on other sites

  • Advertisement