Do Not Trust Additive Explanations

Alicja Gosiewska, Przemyslaw Biecek
3/27/2019
cs.LGstat.ML

Official code: ModelOriented/iBreakDown (84 stars, GPL-3.0, last commit 2023-11-30).

Abstract

Explainable Artificial Intelligence (XAI)has received a great deal of attention recently. Explainability is being presented as a remedy for the distrust of complex and opaque models. Model agnostic methods such as LIME, SHAP, or Break Down promise instance-level interpretability for any complex machine learning model. But how faithful are these additive explanations? Can we rely on additive explanations for non-additive models? In this paper, we (1) examine the behavior of the most popular instance-level explanations under the presence of interactions, (2) introduce a new method that detects interactions for instance-level explanations, (3) perform a large scale benchmark to see how frequently additive explanations may be misleading.

View on arXivView PDF

Code Implementations(1)

ModelOriented/iBreakDownOfficial100%
GPL-3.0

Break Down with interactions for local explanations (SHAP, BreakDown, iBreakDown)

8416R, CSSOct 17, 20182 years agoGPL-3.0
breakdownimlinterpretabilityshapleyxai

Similar Papers

Unified Explanations in Machine Learning Models: A Perturbation Approach

86% match1 repo2024

Discriminative Feature Attributions: Bridging Post Hoc Explainability and Inherent Interpretability

83% match2023

"Why Should I Trust You?": Explaining the Predictions of Any Classifier

83% match1 repo3232016

Counterfactual-based Saliency Map: Towards Visual Contrastive Explanations for Neural Networks

83% match2023

SAILS: Surrogate-based Analysis of Interactions via Local Effect Smooths

83% match1 repo2026

Evaluating Readability and Faithfulness of Concept-based Explanations

83% match1 repo52024

Cite this paper

@article{gosiewska2019trust,
  title  = {Do Not Trust Additive Explanations},
  author = {Alicja Gosiewska and Przemyslaw Biecek},
  year   = {2019},
  eprint = {1903.11420},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1903.11420}
}

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/1903.11420.svg)](https://codeofpaper.com/paper/1903.11420)

Discussion