The University of Amsterdam (UvA) participated in Supercomputing 2024 in Atlanta, where researchers presented their work on using microservices for collaborative model training on the FABRIC testbed. The UvA team, including Alexandros Koufakis, Anestis Dalgitsis, Cees de Laat, and Paola Grosso, demonstrated how vertical federated learning techniques can ensure privacy while training AI models.
At Supercomputing 2024, UvA researchers showcased innovative work on collaborative AI model training. The focus was on vertical federated learning techniques, enabling organizations with diverse datasets to jointly train models without sharing sensitive data.
On Monday, November 18, the team presented their paper at the INDIS 2024 workshop. Titled “Microservices-based FABRIC Implementation of Collaborative Model Training Scenario with Vertical Federated Learning,” it addressed challenges in using geographically distributed, proprietary data. Leveraging the FABRIC infrastructure and DYNAMOS middleware, the team demonstrated a use case for predicting building energy consumption, inspired by collaborations with Prets. This approach supports sustainable energy adoption and facilitates renovation financing.
On Tuesday, November 19, the team showcased a live demo at the SCinet Theater. The demonstration illustrated how the FABRIC network and Vertical Federated Learning can predict energy consumption while preserving data privacy.
From Monday to Thursday, the team was hosted at CIENA’s booth, further demonstrating their work and engaging with visitors from around the world.
UvA extends its gratitude to CIENA and Prets for supporting these presentations and demonstrations.
Read also
