Using attribution methods, we use attributions that filter outirrelevant parts of the input features and then verify the effectiveness of this approach by measuring the classification accuracy of a pre-trained DNN . We also provide Gradient-based attribution (GBA) methods to obtain the attribution mask that further improves the accuracy . As an example, we achieve the accuracy ranging from 99.8\% to 99.9\% on CIFAR-10 using attribution mask obtained from GxSI . We achieve theaccuracy ranging from 0.8 \,*\,Sign-of-Input (GxSI) to 0.9% without additional training. We also provided Gradient*,*,Sign of-Of-Output (GXSI) tools that further improve the accuracy of the attribution masks that further improved the accuracy. We achieved the

Author(s) : Jae-Hong Lee, Joon-Hyuk Chang

Links : PDF - Abstract

Code :
Coursera

Keywords : attribution - accuracy - mask - gxsi - achieve -

Leave a Reply

Your email address will not be published. Required fields are marked *