Abstract
With Artificial Intelligence and Machine Learning (ML) on the rise, organisations of different scales and nature are looking to utilise ML systems to support their day-to-day operations. Many enterprises find it difficult to adapt existing ML solutions to their organisations without huge investments in solution understanding, customisation, infrastructure enablement and workforce training. Some organisations utilise external service providers to provision their standard analytics services, and this often leads to solutions that either do not fit well with their organisation goals or may lead to the loss of expert knowledge behind the establishment of the AI system. This paper aims to address some of these challenges by proposing an ontology for ensuring the reproducibility of ML models in research as well as their integration within application environments. Our work will ensure that the knowledge about a developed ML system or process is accumulated and recorded within an organisation and can be used in the future, either by new employees or other teams within the organisation. This approach can also be utilised by researchers and developers of ML systems to record and publish metadata of their studies, ensuring that future researchers can reuse their work with minimal effort.
Recommended Citation
Bandara, Madhushi; Jiang, Yuchao; Gill, Asif; Rabhi, Fethi A.; and Beydon, Ghassan, "An Application Ontology for Reproducibility of Machine Learning Solutions" (2023). ACIS 2023 Proceedings. 57.
https://aisel.aisnet.org/acis2023/57