SYNTHEMA consortium partners publish SMPC federated learning app on Flower Hub

The SYNTHEMA project consortium continues its work on tools that support privacy-preserving research. As part of these efforts, partners have contributed to the publication of a federated learning application on Flower Hub, developed within the framework of the project.

The application, released by Netcompany RID, implements a peer-to-peer Secure Multi-Party Computation protocol for federated learning using the Flower federated learning framework. It is available through Flower Hub, a platform designed for publishing and running federated learning applications across different environments.

Federated learning enables decentralised training of machine learning models without sharing raw data. In conventional approaches, a central server is typically required to aggregate model updates. In this implementation, aggregation is supported through an additive secret-sharing-based protocol, allowing participants to contribute to model training while maintaining the confidentiality of their individual updates.

The application relies on Flower’s Messages API to coordinate the exchange of encrypted shares between participants. Each client splits its model updates into multiple shares and distributes them to other participants. These shares are then recombined locally, allowing aggregation to take place without exposing individual contributions to a central entity. The server acts as a relay for message exchange and performs weighted averaging on already aggregated results.

The system is designed to operate both in simulation environments and in deployment settings. It supports configurable training rounds and client participation, allowing adaptation to different experimental and operational requirements. This flexibility enables testing under controlled conditions as well as implementation in distributed infrastructures.

The application contributes to SYNTHEMA’s work on secure and distributed approaches to health data analysis in the context of rare haematological diseases.

The work leading to the publication involved contributions from Sofia Tsekeridou, Petros Demetrakopoulos, and Themistoklis Anagnostopoulos. The development also benefited from collaboration with Dimitris Stripelis and Pedro Mesa from Flower Labs.

The application is available on Flower Hub and can be accessed here: https://flower.ai/apps/synthema/smpc-fl/