6950aa02ae8613af620668146dd11840-AuthorFeedback.pdf
–Neural Information Processing Systems
This loss is by definition normalized by theR size, thus the size ofR doesn'tdirectly46 influence the optimization convergence. In Figure 1c we show the effect ofRsize, where saliencies for eachRwere47 obtainedwiththesameoptimiz.
Neural Information Processing Systems
Feb-12-2026, 11:36:46 GMT