RISE: Randomized Input Sampling for Explanation of Black-box Model

Abstract

- Deep neural networks are prevalent and successful, but the decision making process is difficult to explain

- this paper proposes RISE, which generates an importance map indicating how salient each pixel is for the model's prediction

- it is a black-boxed model considering the deep learning model involved

- it has good generality and outperforms most of the existing methods

3 Randomized Input Sampling for Explanation (RISE)