Journal article
Federated explainable artificial intelligence (fXAI): a digital manufacturing perspective
International journal of production research, Vol.62(1-2), pp.171-182
01/17/2024
DOI: 10.1080/00207543.2023.2238083
Abstract
The industry has embraced digitalisation leading to a greater reliance on models derived from data. Understanding and getting insights into the models generated by machine learning algorithms is a challenge due to their non-explicit nature. Explainable artificial intelligence (XAI) is to enhance understanding of the digital models and confidence in the results they produce. The paper makes two contributions. First, the XRule algorithm proposed in the paper generates explicit rules meeting user's preferences. A user may control the nature of the rules generated by the XRule algorithm, e.g. degree of redundancy among the rules. Second, in analogy to federated learning, the concept of federated explainable artificial intelligence (fXAI) is proposed. Besides providing insights into the models built from data and explaining the predicted decisions, the fXAI provides additional value. The user-centric knowledge generated in support of fXAI may lead to discovery of previously unknown parameters and subsequently models that may benefit the non-explicit and explicit perspectives. The insights from fXAI could translate into new ways of modelling the phenomena of interest. A numerical example and three industrial applications illustrate the concepts presented in the paper.
Details
- Title: Subtitle
- Federated explainable artificial intelligence (fXAI): a digital manufacturing perspective
- Creators
- Andrew Kusiak - The University of Iowa
- Resource Type
- Journal article
- Publication Details
- International journal of production research, Vol.62(1-2), pp.171-182
- Publisher
- Taylor & Francis
- DOI
- 10.1080/00207543.2023.2238083
- ISSN
- 0020-7543
- eISSN
- 1366-588X
- Language
- English
- Electronic publication date
- 07/22/2023
- Date published
- 01/17/2024
- Academic Unit
- Nursing; Industrial and Systems Engineering
- Record Identifier
- 9984446289802771
Metrics
4 Record Views