Collective self-learning by exchanging ML models

Collective self-learning based on Machine Learning (ML) model sharing and combination is proposed to accelerate ML-based algorithm deployment. The considered architecture is presented, together with different alternatives for combining ML models. Performance analysis is carried out on an illustrative use case for autonomic optical transmission. The research leading to these results has received funding from the AEI/FEDER TWINS project (TEC2017-90097-R), from the EC METRO-HAUL project (G.A. nº 761727), and from the Catalan ICREA Institution.