Analytics Feature Space: a Novel Framework for Interoperable Edge Machine Learning Detection

Linh Truong, Nguyen Ngoc Nhu Trang

Tutkimustuotos: TyöpaperiEsipainosScientific

29 Lataukset (Pure)

Abstrakti

Departure from the analytics for single assets, in many application domains, complex scenarios require us to determine and discover facts and insights for a collective of assets, spaces, and environments. Due to the advance of machine learning (ML), such complex scenarios are increasingly relied on different edge ML detection, which are diverse for different purposes, like object detection, anomaly detection, surface defect detection or activity recognition/classification, in a unified management view of a business. Widely integrated and provisioned at the edge, the diversity of the underlying ML models, concrete deployments, and changes in operations lead to different types of detection results. However, extensive metadata required for interpreting the results is not well supported, in addition to the difficulty when integrating and analyzing detection results from multiple pipelines and algorithms. In this paper, we present Analytics Feature Space (AFS) as a novel framework to support high-level analytics and integration for multiple types of ML detection at the edge in a unified way that is applied to collectives of assets, spaces and environments. AFS abstracts and supports key metadata related to edge ML detection, changes in detection deployments and metadata for the above-mentioned collective detection and analytics. We introduce techniques to manage relationships between analytics subjects, ML detection models and results. Different ways for integrating AFS and detection pipelines are presented. We demonstrate our experiments with realistic scenarios for operation management.
AlkuperäiskieliEnglanti
TilaJätetty - 29 maalisk. 2024
OKM-julkaisutyyppiEi oikeutettu

Sormenjälki

Sukella tutkimusaiheisiin 'Analytics Feature Space: a Novel Framework for Interoperable Edge Machine Learning Detection'. Ne muodostavat yhdessä ainutlaatuisen sormenjäljen.

Siteeraa tätä