Recently, I've found good success in truncated neural networks ie functions of the form $$ g=f1_{[-M,M]^d}, $$ where $f:\mathbb{R}^d\to\mathbb{R}^n$ is a feed-forward neural network and $1_{[-M,M]^d}$ is the indicator function on the cube of radius $M>0$.

Has anyone come across any paper using these "truncated neural networks" instead of simply using (un-truncated/classical) feed-forward neural networks?

  • $\begingroup$ Hi and welcome to this community! It's not clear how the indicator function is supposed to select parts of the neural network. Can you clarify this? Where did you see this notation? $\endgroup$ – nbro Jan 15 at 13:15
  • $\begingroup$ A colleague of mine was using them. It acts but muliplication and M is tuned separately. $\endgroup$ – AIM_BLB Jan 15 at 20:35
  • $\begingroup$ I still don't understand exactly how the truncation of the NN is supposed to work. Can you clarify this? How is M supposed to truncate the NN? I understood that M is learned. $\endgroup$ – nbro Jan 16 at 0:36
  • $\begingroup$ Yes but the function g (truncated network)'s outputs area all set to zero when the input is larger than M $\endgroup$ – AIM_BLB Jan 16 at 9:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.