11-20, 10:20–10:25 (Pacific/Auckland), WG403
Digital Earth is implementing a Machine Learning Operations proof of concept for an automated system designed to support development, training, and release of complex ML models for satellite imagery such as artificial surface detection. Talk will share architecture and learnings, focusing on integration of open-source components with cloud services.
Digital Earth (DE) is implementing a Machine Learning Operations proof-of-concept for an automated system designed to support the end-to-end development, training, and release of complex machine learning workflows and models for satellite imagery, such as artificial surface detection used in Land Cover. The PoC focuses on the discoverability of open-source machine learning models, streamlining data versioning, feature transformation, containerised model training, hyperparameter tuning, governance, model, and artefact management. The lightening talk will share our architecture and learnings, with a focus on the integration of various open-source components and specifications with cloud-services.
James is a Technical Lead in Research and Product Development and is passionate about improving the ability to detect change and create insights in Australia’s landscape using earth observation data. He enjoys taking scientific research and applying operational large scale data engineering techniques to solve national problems. His work includes designing and developing geospatial scientific products, data quality improvements, big data engineering pipelines, cloud infrastructure and customer facing platforms and applications.