Deep Integrated Explanations

Oren Barkan, Yehonatan Elisha, Jonathan Weill, Yuval Asher, Amit Eshel +1 more
10/23/2023
cs.CV

No official code found. Best implementation: yonisGit/dix (3 stars, last commit 2026-03-15).

Abstract

This paper presents Deep Integrated Explanations (DIX) - a universal method for explaining vision models. DIX generates explanation maps by integrating information from the intermediate representations of the model, coupled with their corresponding gradients. Through an extensive array of both objective and subjective evaluations spanning diverse tasks, datasets, and model configurations, we showcase the efficacy of DIX in generating faithful and accurate explanation maps, while surpassing current state-of-the-art methods.

View on arXivView PDF

Code Implementations(1)

yonisGit/dixCommunity100%

PyTorch Implementation of Deep Integrated Explanations.

30PythonAug 29, 20237 months ago

Similar Papers

Visual Explanations via Iterated Integrated Attributions

92% match1 repo152026

Learning to Explain: A Model-Agnostic Framework for Explaining Black Box Models

88% match1 repo32023

Language Models Can Explain Visual Features via Steering

88% match2026

GIFT: A Framework Towards Global Interpretable Faithful Textual Explanations of Vision Classifiers

87% match2024

Language Model as Visual Explainer

86% match2026

LeGrad: An Explainability Method for Vision Transformers via Feature Formation Sensitivity

86% match1 repo992024

Cite this paper

@article{barkan2023deep,
  title  = {Deep Integrated Explanations},
  author = {Oren Barkan and Yehonatan Elisha and Jonathan Weill and Yuval Asher and Amit Eshel and Noam Koenigstein},
  year   = {2023},
  eprint = {2310.15368},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2310.15368v2}
}

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/2310.15368.svg)](https://codeofpaper.com/paper/2310.15368)

Discussion