Comparison of Contextual Importance and Utility with LIME and Shapley Values

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

*Tämän työn vastaava kirjoittaja

Tutkimustuotos: Artikkeli kirjassa/konferenssijulkaisussaConference contributionScientificvertaisarvioitu

Abstrakti

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.

AlkuperäiskieliEnglanti
OtsikkoExplainable and Transparent AI and Multi-Agent Systems - 3rd International Workshop, EXTRAAMAS 2021, Revised Selected Papers
ToimittajatDavide Calvaresi, Amro Najjar, Michael Winikoff, Kary Främling
KustantajaSpringer Science and Business Media Deutschland GmbH
Sivut39-54
Sivumäärä16
ISBN (painettu)9783030820169
DOI - pysyväislinkit
TilaJulkaistu - 2021
OKM-julkaisutyyppiA4 Artikkeli konferenssijulkaisuussa
TapahtumaInternational Workshop on Explainable, Transparent AI and Multi-Agent Systems - Virtual, Online
Kesto: 3 toukokuuta 20217 toukokuuta 2021
Konferenssinumero: 3

Julkaisusarja

NimiLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
KustantajaSpringer
Vuosikerta12688 LNAI
ISSN (painettu)0302-9743
ISSN (elektroninen)1611-3349

Workshop

WorkshopInternational Workshop on Explainable, Transparent AI and Multi-Agent Systems
LyhennettäEXTRAAMAS
KaupunkiVirtual, Online
Ajanjakso03/05/202107/05/2021

Sormenjälki

Sukella tutkimusaiheisiin 'Comparison of Contextual Importance and Utility with LIME and Shapley Values'. Ne muodostavat yhdessä ainutlaatuisen sormenjäljen.

Siteeraa tätä