hh.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
EcoShap: Save Computations by only Calculating Shapley Values for Relevant Features
Halmstad University, School of Information Technology, Center for Applied Intelligent Systems Research (CAISR).ORCID iD: 0000-0001-7055-2706
Halmstad University, School of Information Technology, Center for Applied Intelligent Systems Research (CAISR).ORCID iD: 0000-0002-7796-5201
Halmstad University, School of Information Technology, Center for Applied Intelligent Systems Research (CAISR).ORCID iD: 0000-0003-2590-6661
2024 (English)In: Artificial Intelligence. ECAI 2023 International Workshops: XAI^3, TACTIFUL, XI-ML, SEDAMI, RAAIT, AI4S, HYDRA, AI4AI, Kraków, Poland, September 30 – October 4, 2023, Proceedings, Part I / [ed] Nowaczyk, Sławomir et al., Cham: Springer, 2024, Vol. 1947, p. 24-42Conference paper, Published paper (Refereed)
Abstract [en]

One of the most widely adopted approaches for eXplainable Artificial Intelligence (XAI) involves employing of Shapley values (SVs) to determine the relative importance of input features. While based on a solid mathematical foundation derived from cooperative game theory, SVs have a significant drawback: high computational cost. Calculating the exact SV is an NP-hard problem, necessitating the use of approximations, particularly when dealing with more than twenty features. On the other hand, determining SVs for all features is seldom necessary in practice; users are primarily interested in the most important ones only. This paper introduces the Economic Hierarchical Shapley values (ecoShap) method for calculating SVs for the most crucial features only, with reduced computational cost. EcoShap iteratively expands disjoint groups of features in a tree-like manner, avoiding the expensive computations for the majority of less important features. Our experimental results across eight datasets demonstrate that the proposed technique efficiently identifies top features; at a 50% reduction in computational costs, it can determine between three and seven of the most important features. © The Author(s) 2024.

Place, publisher, year, edition, pages
Cham: Springer, 2024. Vol. 1947, p. 24-42
Series
Communications in Computer and Information Science, ISSN 1865-0929, E-ISSN 1865-0937 ; 1947
Keywords [en]
Explainable Artificial Intelligence (XAI), Feature Importance, Shapley Value
National Category
Computer Sciences
Identifiers
URN: urn:nbn:se:hh:diva-52744DOI: 10.1007/978-3-031-50396-2_2Scopus ID: 2-s2.0-85184111581ISBN: 978-3-031-50395-5 (print)ISBN: 978-3-031-50396-2 (electronic)OAI: oai:DiVA.org:hh-52744DiVA, id: diva2:1840492
Conference
International Workshops of the 26th European Conference on Artificial Intelligence (ECAI 2023), Kraków, Poland, 30 September-4 October, 2023
Funder
Swedish Research Council, CHIST-ERA-19-XAI-012Available from: 2024-02-23 Created: 2024-02-23 Last updated: 2024-12-04Bibliographically approved
In thesis
1. Towards better XAI: Improving Shapley Values and Federated LearningInterpretability
Open this publication in new window or tab >>Towards better XAI: Improving Shapley Values and Federated LearningInterpretability
2024 (English)Licentiate thesis, comprehensive summary (Other academic)
Abstract [en]

The use of Artificial Intelligence (AI) in various areas has resulted in significant progress. However, it has also raised concerns about how transparent, understandable, and reliable AI models are. Explainable Artificial Intelligence(XAI) has become an important area of study because it makes AI systems easier for people to understand. XAI also aims to create trust, accountability, and transparency in AI systems, especially in vital areas such as healthcare, finance, and autonomous systems. Furthermore, XAI helps detect biases, improve model debugging, and enable the discovery of new insights from data.

With the increasing focus on the XAI field, we have developed a framework called Field’s Evolution Graph (FEG) to track the evolution of research within the XAI field. This framework helps researchers identify key concepts and their interrelationships over time. Different approaches have been developed in XAI, and Shaply values are among the most well-known techniques. We further examine this method by evaluating its computational cost and analyzing various characteristic functions. We have introduced EcoShap, a computationally efficient method for calculating Shapley values to identify the most important features. By focusing calculations on a few of the most important features, EcoShap significantly reduces the computational cost, making it feasible to apply Shapley values to large datasets.

The thesis is extended by analyzing different characteristic functions used in Shapley value calculations theoretically and practically. We examine how feature importance rankings are reliable and how considering various characteristic functions, like accuracy and F1-score metrics, affects those rankings.

Additionally, Federated Learning (FL) is a machine learning paradigm that is able to train global models from different clients while keeping data decentralized. Similar to other machine learning paradigms, XAI is needed in this context. To address this need, we have proposed a method to use Incremental Decision Trees as an inherently interpretable model within the FL framework, offering an interpretable alternative to black-box models. The objective is to employ inherently explainable models instead of trying to explain black-box models. Three aggregation strategies have been presented to combine local models into a global model while maintaining interpretability and accuracy.

In summary, the research in this thesis contributes to the field of XAI by introducing new methods to improve efficiency, analyzing existing methods to assess the reliability of XAI techniques, and proposing a solution for utilizing more intrinsically explainable models in the Federated Learning framework.

Place, publisher, year, edition, pages
Halmstad: Halmstad University Press, 2024. p. 28
Series
Halmstad University Dissertations ; 124
Keywords
eXplainable AI, Shapley Values, Federated Learning
National Category
Computer Systems
Identifiers
urn:nbn:se:hh:diva-54994 (URN)978-91-89587-65-6 (ISBN)978-91-89587-64-9 (ISBN)
Presentation
2025-01-08, S3030, Kristian IV:s väg 3, Halmstad, 13:00 (English)
Opponent
Supervisors
Funder
VinnovaKnowledge Foundation
Available from: 2024-12-04 Created: 2024-12-04 Last updated: 2024-12-04Bibliographically approved

Open Access in DiVA

No full text in DiVA

Other links

Publisher's full textScopus

Authority records

Jamshidi, ParisaNowaczyk, SławomirRahat, Mahmoud

Search in DiVA

By author/editor
Jamshidi, ParisaNowaczyk, SławomirRahat, Mahmoud
By organisation
Center for Applied Intelligent Systems Research (CAISR)
Computer Sciences

Search outside of DiVA

GoogleGoogle Scholar

doi
isbn
urn-nbn

Altmetric score

doi
isbn
urn-nbn
Total: 110 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf