Comparison of Contextual Importance and Utility with LIME and Shapley Values

Kary Främling*, Marcus Westberg, Martin Jullum, Manik Madhikermi, Avleen Malhi

*Corresponding author for this work

Research output: Chapter in Book/Conference proceedings/Edited volumeConference contributionScientificpeer-review

12 Citations (Scopus)

Abstract

Different explainable AI (XAI) methods are based on different notions of ‘ground truth’. In order to trust explanations of AI systems, the ground truth has to provide fidelity towards the actual behaviour of the AI system. An explanation that has poor fidelity towards the AI system’s actual behaviour can not be trusted no matter how convincing the explanations appear to be for the users. The Contextual Importance and Utility (CIU) method differs from currently popular outcome explanation methods such as Local Interpretable Model-agnostic Explanations (LIME) and Shapley values in several ways. Notably, CIU does not build any intermediate interpretable model like LIME, and it does not make any assumption regarding linearity or additivity of the feature importance. CIU also introduces the value utility notion and a definition of feature importance that is different from LIME and Shapley values. We argue that LIME and Shapley values actually estimate ‘influence’ (rather than ‘importance’), which combines importance and utility. The paper compares the three methods in terms of validity of their ground truth assumption and fidelity towards the underlying model through a series of benchmark tasks. The results confirm that LIME results tend not to be coherent nor stable. CIU and Shapley values give rather similar results when limiting explanations to ‘influence’. However, by separating ‘importance’ and ‘utility’ elements, CIU can provide more expressive and flexible explanations than LIME and Shapley values.
Original languageEnglish
Title of host publicationExplainable and Transparent AI and Multi-Agent Systems - 3rd International Workshop, EXTRAAMAS 2021, Revised Selected Papers
EditorsDavide Calvaresi, Amro Najjar, Michael Winikoff, Kary Främling
PublisherSpringer
Pages39-54
Number of pages16
ISBN (Print)9783030820169
DOIs
Publication statusPublished - 2021
Externally publishedYes
Event3rd International Workshop on Explainable, Transparent AI and Multi-Agent Systems, EXTRAAMAS 2021 - Virtual, Online
Duration: 3 May 20217 May 2021

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume12688 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference3rd International Workshop on Explainable, Transparent AI and Multi-Agent Systems, EXTRAAMAS 2021
CityVirtual, Online
Period3/05/217/05/21

Keywords

  • Contextual Importance and Utility
  • Explainable AI
  • Outcome explanation
  • Post hoc explanation

Fingerprint

Dive into the research topics of 'Comparison of Contextual Importance and Utility with LIME and Shapley Values'. Together they form a unique fingerprint.

Cite this