MLEDGE: A sustainable initiative for the development of efficient networks and solutions in the Cloud and Federated Learning
IMDEA Networks/DICYT Data-driven decision-making driven by Machine Learning (ML) algorithms is changing the way society and the economy work, with a profound positive impact on our daily lives. For these solutions to be effective, data often has to be processed close to the end user and such data can be private and confidential.
Distributed and, in particular, Federated Learning (FL) emerges as a leading paradigm within the ML branch satisfying these two properties. Federated Learning has grown in parallel with the expansion of the cloud towards the edge (CloudEdge) but, interestingly, the two have developed mostly independently, despite their natural parallelism. In MLEDGE (Cloud and Edge Machine Learning), IMDEA Networks will work, with Dr. Nikolaos Laoutaris as principal investigator, to reverse this trend by implementing FL as an independent but optimized cross-sector layer on top of Cloud Edge, with real-world applications and data to demonstrate that this synergy can yield great benefits for all.
It is estimated that the data economy will reach an impact of €827 billion for the 27 EU countries by 2025 (according to the European data strategy for the period 2025-2030). Therefore, the goal is to enable a thriving ecosystem of secure and efficient FL services at the edge, capable of facilitating the use of sensitive personal and B2B data to train machine learning models (for individual end-users or administratively independent organizations collaborating under different trust assumptions – from full to zero, and any level in between).
Efficiency, sustainability and security
As highlighted by Elisa Cabana, Postdoc Researcher at IMDEA Networks: “The project contributes research, among other areas, in federated learning as a Service (FLaaS), Cloud Edge processing, efficient use of FL in hybrid clouds and protection against attacks, protection of sensitive or confidential data being exchanged, management of data portability challenges at the edge, etc.”. In this context, the team will design a development framework and components to help popularize this type of services, as well as solutions against poisoning or inference attacks launched from unruly edge servers and/or “honest but curious” aggregation nodes. It includes the creation of a ‘watermark’ to protect against redistribution of data or metadata exchanged between servers at the edge under FLaaS.
Other highlights will be, as Cabana summarizes, “Creating an economic and business logic layer that implements a fair distribution of costs and revenues between parties when collaborating on ML model training, and supporting DevOps (A set of practices that bring together software development and IT operations. Its goal is to speed up the software development life cycle and provide high-quality continuous delivery) and the continued development of machine learning services in the cloud, optimizing costs by monitoring, predicting and intelligently and energy-efficiently allocating computational jobs.” The research will further contribute to design, implement and make public demonstrators that work with sensitive data of individuals and feed useful models in areas of the traditional and digital economy such as FinTech, identity, health, transportation, access control, etc.
Technology transfer to society
The innovation of the project will develop, in an international context, favorable market conditions for the use of federated learning in the cloud and in federated data architectures, such as, for example, those defined by institutions like IDSA or Gaia-X, with innovations of great interest to address important economic, business and social problems linked to the existence of silos in the storage and exploitation of data in the economy. “MLEDGE will make advanced federated learning technology accessible to more organizations and individuals, including SMEs and government agencies, and will favor the creation of sustainable businesses for all actors in the value chain (machine learning experts/suppliers, cloud and data service providers, traditional and digital industry, public sector and academia, etc.)”, says Nikolaos Laoutaris, Research Professor at IMDEA Networks and Principal Investigator of the project at the institute.
The project will be fundamental for the development of Cloud and ML/FL infrastructures in Spain and for the promotion of national R&D&I. It will contribute to the Sustainable Development Goals set by the United Nations for 2030 and will promote the sustainable development of efficient networks and FL solutions through practical work that can substantially and positively impact the environment. In terms of technological solutions, the following can be highlighted:
- Traditional economics (construction, finance, health, etc.). The use case will be developed by a company to improve processes, or decision making (e.g., in real time) based on data or models coming from FL.
- Digital economy. One example could be in the field of digital health, such as exploiting information from mobile devices or wearable technologies. Another would be the training of digital advertising models.
- Optimization of CloudEdge infrastructures. A key functionality for MLEDGE, which will use machine learning algorithms trained in a federated way.
MLEDGE (Cloud and Edge Machine Learning) has funding (January 2023-June 2025) from the Ministry of Economic Affairs and Digital Transformation, European Union NextGeneration-EU.