Projects

GenF: Federated Learning as a Service (FLaaS) with Generative AI

Generative AI to improve Federated Learning

Funding: ICSC Innovation Funds

Enabling technology: Artificial Intelligence, Machine Learning

Sustainable Development Goals

GenF aims to extend FLaaS by incorporating generative AI features to improve privacy and model training in FL. Peers within the federation initially train only on synthetic data, and knowledge exchange uses evidence-based methods. The privacy-preserving synthesis process involves navigating the latent spaces of generative models to obfuscate sensitive data. This innovation is designed to address the security gaps of conventional FL systems, offering a privacy-focused solution for healthcare applications.

National Centre for HPC, Big Data and Quantum Computing (ICSC), a project funded by the European Union – NextGenerationEU – and by Italy’s National Recovery and Resilience Plan (PNRR) – Mission 4, Component 2.

Objectives

The main goal is to extend FLaaS by improving peer-to-peer learning capabilities. This goal can be achieved by leveraging the transformative potential of generative AI to promote a distributed learning strategy that is both privacy-preserving and data-driven.

Initial challenge

Improving the state of the art in privacy-preserving generative AI and FL. After gaining an overview of the most effective methods for privacy-preserving data synthesis, the project aims to integrate generative AI into the FLaaS prototype.

Solution

The integration of generative AI results in an advanced FLaaS prototype: this will give the prototype improved and more effective capabilities for privacy-preserving distributed learning.

Benefits

Some expected outcomes are:

Contributing to the broader scientific community’s understanding of FLaaS’s potential applications and benefits in healthcare.
Advancing the vision of federated learning in clinical settings.
Promoting collaborative progress in privacy-preserving machine learning.

For industry, the improved FLaaS prototype is poised to position itself as a pioneering force in the federated learning market. The goal is to engage clinical centers in federated learning without directly sharing sensitive patient data, thereby fostering the creation of more robust and accurate predictive models.

IFAB’s role

IFAB took part in the project in an operational capacity, contributing to the management and coordination of activities across the different development phases. This involvement covered both organizational aspects and liaison among the parties involved, ensuring consistency between the activities carried out and the project’s objectives. Through this contribution, IFAB helped ensure the project’s proper execution and the delivery of the expected results, making its management expertise available in support of the entire process.