Comment on Neural Network Follies (2003)parentComments−albertzeyer4yYou 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.−pietro72ohboy4yBoth 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.−danuker4yYour approach will compute pixel importance in a single stepHow do you address the vanishing gradient problem?
Comments
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.
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.
How do you address the vanishing gradient problem?