CeADAR at ECML PKDD 2025: Showcasing European AI Research in Porto

CeADAR recently had the opportunity to represent three European research projects – ICOS, MANOLO, and O-CEI – at the European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD) in Porto presenting two research pieces, OASIS and DriftMoE.

OASIS is a framework created as the Intelligence Layer for the ICOS project. The Intelligence LayerModule is a key component of the ICOS platform, designed to enable AI-driven optimization, predictive analytics, and collaborative model sharing across the edge-cloud continuum. Acting as the interface between the Meta-Kernel and User layers, it orchestrates the full lifecycle of machine learning models from training and inference to monitoring and explainability. 

OASIS has been developed over the last three years by the CeADAR team members: Jaydeep Samanta, Sebastián Cajas, Romila Ghosh, Dr. Andrés L. Suárez-Cetrulo, and Dr. Ricardo Simón Carbajo. Funded by the HORIZON EU.

DriftMoE, is a novel framework addressing a common challenge when deploying models in production,  concept drift, where data patterns evolve over time making models less accurate. This paper is a collaborative effort of the MANOLO, O-CEI, and ICOS projects. It is a research topic that is core to MANOLO project, as the proposed algorithm addresses model robustness with continual model training.

DriftMoE uses an online Mixture of Experts (MoE) approach with a neural router and co-training strategies to adapt quickly to changing data streams.