Institutional Repository of Vilnius University
Paaiškinamojo dirbtinio intelekto (PDI) pateikiamo paaiškinimo vertinimo metodų unifikacija /
Abstract
dc:descriptionThe relevence and popularity of artificial intelligence (AI) is growing every year, yet due to the lack of trust and transparancy the adoption of AI applications in high stake situations (e. g. law, medicine, etc.) remains underwhelming. To tackle these issues, solutions based on explainable artificial intelligence (XAI) started to get developed, yet despite its recent popularity it often remains unclear whether an explanation satisfies the needs of the users and often whether it’s a correct explanation at all. Due to these reasons there is a necessity to evaluate explanations using explanation evaluation methods. However, due to the XAI field being in its primary stages of development, the current evaluation methods have multiple flaws. This paper analyses four of the evaluation methods which are found in the scientific literature by examining their core principles and identifying their weaknesses. The result of this analysis is a newly formed 1-CoRE evaluation method, which was created by unifying the strengths of the aformentioned methods and by tackling the more problematic areas, which were identified throughout the analysis.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Grumodas, Kęstutis,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD146241282&prefLang=en_US