Using Stratified Sampling to Improve LIME Image Explanations

Muhammad Rashid, Elvio G. Amparore, Enrico Ferrari, Damiano Verda
3/26/2024
cs.AI

No official code found. Best implementation: rashidrao-pk/lime_stratified (1 star, BSD-2-Clause, last commit 2024-10-07). 1 more below.

Abstract

We investigate the use of a stratified sampling approach for LIME Image, a popular model-agnostic explainable AI method for computer vision tasks, in order to reduce the artifacts generated by typical Monte Carlo sampling. Such artifacts are due to the undersampling of the dependent variable in the synthetic neighborhood around the image being explained, which may result in inadequate explanations due to the impossibility of fitting a linear regressor on the sampled data. We then highlight a connection with the Shapley theory, where similar arguments about undersampling and sample relevance were suggested in the past. We derive all the formulas and adjustment factors required for an unbiased stratified sampling estimator. Experiments show the efficacy of the proposed approach.

View on arXivView PDF

Code Implementations(2)

rashidrao-pk/lime_stratifiedCommunity100%
BSD-2-Clause

Branch of LIME adding stratified Sampling

10CSS, TeXDec 11, 20232 years agoBSD-2-Clause
rashidrao-pk/lime-stratified-examplesCommunity100%
MIT

Examples used to generate plots for paper 'Using stratified sampling to improve LIME Image Explanations' in AAAI24

10CSS, HTMLAug 1, 20232 years agoMIT

Similar Papers

Fixing the RANSAC Stopping Criterion

81% match2025

PSCR: Patches Sampling-based Contrastive Regression for AIGC Image Quality Assessment

79% match1 repo62023

Scalable Model-Assisted Multi-Target Estimation in Large Image Collections

79% match2026

Back to the Baseline: Examining Baseline Effects on Explainability Metrics

79% match1 repo22025

Stratified sampling and bootstrapping for approximate Bayesian computation

78% match1 repo32019

Segmentation and Smoothing Affect Explanation Quality More Than the Choice of Perturbation-based XAI Method for Image Explanations

78% match1 repo2024

Cite this paper

@article{rashid2024using,
  title  = {Using Stratified Sampling to Improve LIME Image Explanations},
  author = {Muhammad Rashid and Elvio G. Amparore and Enrico Ferrari and Damiano Verda},
  year   = {2024},
  eprint = {2403.17742},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2403.17742v1}
}

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/2403.17742.svg)](https://codeofpaper.com/paper/2403.17742)

Discussion