The Federated Learning Framework is one of the most promising products developed by Strategy & Innovation - R&D Spain. The Trustworthy AI team has been developing it for the last two years, leveraging the context of several European projects, such as ALCHIMIA, SEDIMARK, FAME and FERMI The conception of this framework is also connected to ERATOSTHENES project.

What is “Federated Learning”?

Federated Learning is a machine learning technique that allows decentralized AI model training while keeping data local. Clients train local models with their data, sharing only updates with a central server to create a global model. This preserves data privacy, enhances scalability, and complies with regulations. It is used in mobile apps, healthcare, self-driving cars, robotics, and smart manufacturing, enabling efficient, private, and adaptive AI systems.

What makes R&D Spain Federated Learning Framework different from other solutions in the market?

The R&D Spain Federated Learning Framework, developed by the TAI team, distinguishes itself with a Python-based modular and extensible architecture built upon a set of functional units known as pods, which can be combined in numerous ways depending of the specific requirements of each use case. This design allows for flexible, customized solutions adaptable to complex federated learning scenarios, making the framework highly customizable, flexible and adaptable. 

History and prospects

The TAI team initiated the Federated Learning research line a couple of years ago, recognizing its strategic importance due to European Commission funding opportunities. A key requirement for funding is a thorough analysis of the state of the art to highlight the uniqueness and profitability of the proposal, so the team identified a gap in existing open-source frameworks, leading to the development of a tool based on the "pipes and filters" paradigm, with pods and wires abstraction.

Future plans include extending the Federated Learning framework to enable fine-tuning of Large Language Models (LLMs), which is a strategic research line with the advent of Generative AI, and completing the integration of the framework with the MLOps platform being developed by the MLOps team, which is an important next step to industrialize the solution.