D3.2: Constraint-aware ML and analytics services

Description

Τhis deliverable summarizes the research results related to the constraint-aware, adaptive, and scalable machine learning (ML) methods developed in the ExtremeXP. It first introduces novel ML algorithms that embed performance and physics based constraints into loss functions—enhancing cybersecurity classification and flood prediction  ccuracy. It also presents a structure-constrained clustering method (UEC) and a meta learning tool for automated model selection. It then explores online and continual learning for evolving data streams, and a lightweight deployment unit to enable scalable ML workflows across clusters and cloud platforms. Together, these innovations advance robust, efficient, and deployable ML systems. 

Authors

DOI: 10.5281/zenodo.20750093

Publication Date: 2026-06-18

Back to publications list


About