Investigating standards for Machine Learning Training Data
In March 2022, the Open Geospatial Consortium (OGC) opened a call for proposals to be part of the OGC Testbed-18. Testbed-18 provided the opportunity to explore and demonstrate the use of training datasets (TDS) and contribute to the future standardisation of TDS. The scope of this work was to develop an Engineering Report (ER) documenting learnings from the Vicmap Vegetation project, and other Machine Learning (ML) projects in collaboration with the OGC and a UK based EO company, Pixalytics.
This project was of particular interest to FrontierSI, DELWP and NSW as standardisation and re-use of TDS will be essential to the creation and maintenance of spatial datasets, including Foundational Spatial Data (FSD). FrontierSI recently delivered the first state-wide FSDF dataset to DELWP, Vicmap Vegetation, and through this experience identified the need to be able to use and re-use TDS, as well as the ML algorithms. FrontierSI, DELWP and NSW SS collaborated on an adjacent project to review geospatial machine learning (ML) projects previously undertaken by Vicmap and NSW Spatial Services to contribute and learn from the OGC ER.
Partners included OGC, Pixalytics, Curtin University, DELWP (DTP), and NSW Spatial Services.
The OGC Testbed-18 project developed and submitted an Engineering Report to the OGC which was accepted and is publicly available here
The collaboration project developed a series of ML case studies, undertook metadata analysis of the case studies, assessed ML RFP requirements and finally ran some tests and trails of existing data against proposed ML TDS standards. The collaboration delivered a recommendations report which can be used by DELWP and NSW SS when undertaking future ML projects.
The Testbed-18 ER outcomes contribute to the knowledge and progress of the global open geospatial community and future development of ML TDS standards.
ML is still an emerging technology and the recommendations from the collaboration project will help Victoria and NSW achieve procurement of cost effective, fit for purpose and re-usable ML and TDS on future projects. This will aid modernisation of FSD.
Contact FrontierSI at email@example.com to find out more.