XRAI: Better Attributions Through Regions

Andrei Kapishnikov, Tolga Bolukbasi, Fernanda Viégas, Michael Terry
6/6/2019
cs.CVstat.ML

Official code: PAIR-code/saliency (996 stars, Apache-2.0, TensorFlow, last commit 2024-03-20).

Abstract

Saliency methods can aid understanding of deep neural networks. Recent years have witnessed many improvements to saliency methods, as well as new ways for evaluating them. In this paper, we 1) present a novel region-based attribution method, XRAI, that builds upon integrated gradients (Sundararajan et al. 2017), 2) introduce evaluation methods for empirically assessing the quality of image-based saliency maps (Performance Information Curves (PICs)), and 3) contribute an axiom-based sanity check for attribution methods. Through empirical experiments and example results, we show that XRAI produces better results than other saliency methods for common models and the ImageNet dataset.

View on arXivView PDF

Code Implementations(1)

PAIR-code/saliencyOfficial100%
TensorFlowApache-2.0

Framework-agnostic implementation for state-of-the-art saliency methods (XRAI, BlurIG, SmoothGrad, and more).

996191Shell, PythonJun 9, 20172 years agoApache-2.0
convolutional-neural-networksdeep-learningdeep-neural-networksig-saliencyimage-recognition+6 more

Similar Papers

AIM: Adversarial Information Masking for Faithfulness Evaluation of Saliency Maps

87% match2026

Classification Metrics for Image Explanations: Towards Building Reliable XAI-Evaluations

87% match1 repo22024

Rethinking gradient weights' influence over saliency map estimation

87% match2022

SE3D: A Framework For Saliency Method Evaluation In 3D Imaging

87% match1 repo2024

Saliency Methods are Encoders: Analysing Logical Relations Towards Interpretation

87% match2024

Why are Saliency Maps Noisy? Cause of and Solution to Noisy Saliency Maps

87% match1 repo342019

Cite this paper

@article{kapishnikov2019xrai,
  title  = {XRAI: Better Attributions Through Regions},
  author = {Andrei Kapishnikov and Tolga Bolukbasi and Fernanda Viégas and Michael Terry},
  year   = {2019},
  eprint = {1906.02825},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1906.02825}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1906.02825.svg)](https://codeofpaper.com/paper/1906.02825)

Discussion