News for hackers. Stuff that matters.
Re:Neural Network Follies (2003) by albertzeyer on Wednesday May 04, 2022 @03:18AM (#31257666)
You don't need to randomly flip pixels in the input image. You can simply backpropagate to it w.r.t. the loss function and modify it based on the gradient to minimize the loss.
Re:Neural Network Follies (2003) by pietro72ohboy on Wednesday May 04, 2022 @04:20AM (#31257996)
Both approaches are correct. The parent is talking about Ablation based importance attribution whereas you’re using gradients to assign importance.

Your approach will compute pixel importance in a single step whereas ablation based approaches generally need many passes.

Re:Neural Network Follies (2003) by danuker on Wednesday May 04, 2022 @05:12AM (#31258249)
> Your approach will compute pixel importance in a single step

How do you address the vanishing gradient problem?