Training machine learning models collaboratively on geographically distributed environments

What is it about?

Private companies (e.g. airlines) often have private datasets that they cannot share externally, due to privacy constraints. Federated Learning (FL) is an approach that enables companies to train a machine learning (ML) model on their private datasets while respecting data restrictions. This results in a higher-quality ML model compared to training on a single dataset. We showcased a FL use case on airline data using DYNAMOS, a microservice-based middleware for data exchange developed by the Uiversity of Amsterdam. Additionally, our use case was geographically distributed across different continents, using FABRIC (an international testbed infrastructure) to emulate the distributed infrastructure of participating companies.

Featured Image