TL;DR:
- MLEDGE is an initiative focused on integrating Federated Learning (FL) with CloudEdge to create secure and efficient FL services.
- The project aims to enable the use of sensitive data for training machine learning models while ensuring privacy and confidentiality.
- MLEDGE will develop a development framework and components to popularize FL services and defend against attacks.
- It will establish an economic layer for fair cost distribution and support DevOps for agile software development.
- The project will showcase real-world applications in areas like FinTech, identity management, and transportation.
- MLEDGE seeks to foster technology transfer, create market conditions for FL adoption, and contribute to sustainable development goals.
Main AI News:
The rapid advancement of data-driven decision-making, fueled by cutting-edge Machine Learning (ML) algorithms, is revolutionizing the fabric of our society and economy, ushering in a wave of positive transformations in our daily lives. However, to ensure the efficacy of these solutions, it is imperative to process data in close proximity to end users, often involving private and confidential information.
In the realm of ML, Distributed and Federated Learning (FL) has emerged as a leading paradigm that satisfies these crucial requirements. While Federated Learning has thrived alongside the expansion of cloud computing towards the edge (CloudEdge), it is remarkable to note that these two domains have predominantly developed independently, despite their inherent parallelism. Enter MLEDGE (Cloud and Edge Machine Learning), a pioneering initiative led by IMDEA Networks and spearheaded by Dr. Nikolaos Laoutaris, which aims to reverse this trend by seamlessly integrating FL as an independent yet optimized cross-sector layer atop CloudEdge. Through real-world applications and tangible data demonstrations, MLEDGE seeks to showcase the tremendous benefits that this synergy can unlock for all stakeholders.
Projections indicate that by 2025, the data economy is poised to generate a staggering impact of €827 billion across the 27 European Union countries (1)*. Against this backdrop, the primary objective of MLEDGE is to foster a thriving ecosystem of secure and efficient FL services at the edge, empowering the utilization of sensitive personal and B2B data for training machine learning models, whether catering to individual end-users or facilitating collaborative efforts among administratively independent organizations operating under varying trust assumptions (ranging from full to zero, and every level in between), MLEDGE endeavors to foster a landscape where efficiency, sustainability, and security harmoniously coexist.
Elisa Cabana, a distinguished Postdoc Researcher at IMDEA Networks, emphasizes the significant contributions of the project: “MLEDGE advances research in various domains, including Federated Learning as a Service (FLaaS), CloudEdge processing, efficient utilization of FL in hybrid clouds, protection against attacks, secure exchange of sensitive and confidential data, management of data portability challenges at the edge, and much more.” Within this context, the MLEDGE team is poised to design a comprehensive development framework and accompanying components that will fuel the widespread adoption of these services.
Moreover, the initiative will focus on developing robust defenses against poisoning or inference attacks launched from unruly edge servers, as well as “honest but curious” aggregation nodes. Of notable significance is the creation of a robust ‘watermark’ mechanism, which effectively safeguards against the unauthorized redistribution of data or metadata exchanged between servers operating under FLaaS.
Among the key highlights of MLEDGE, as succinctly encapsulated by Cabana, is the establishment of an economic and business logic layer. This layer will facilitate the equitable distribution of costs and revenues among collaborating parties engaged in ML model training. Additionally, the initiative aims to provide seamless support for DevOps—a set of practices that synergize software development and IT operations—thereby expediting the software development life cycle and ensuring high-quality continuous delivery. The project’s research endeavors will further culminate in the design, implementation, and public showcasing of demonstrators that operate with sensitive individual data, effectively harnessing it to generate useful models across domains such as FinTech, identity management, healthcare, transportation, access control, and more, encompassing both the traditional and digital economy.
MLEDGE’s impact transcends mere technological advancements; it extends to fostering societal technology transfer. By innovating in an international context, the project is poised to create favorable market conditions for the widespread adoption of federated learning in the cloud and federated data architectures. Notably, these architectures align with the principles set forth by esteemed institutions like IDSA or Gaia-X, and their integration with MLEDGE will present unprecedented opportunities to address critical economic, business, and social challenges tied to the pervasive existence of data silos in our economy.
“MLEDGE will democratize access to advanced federated learning technologies, making them accessible to a diverse range of organizations and individuals, including SMEs and government agencies. It will foster the creation of sustainable businesses for all stakeholders in the value chain, spanning machine learning experts/suppliers, cloud and data service providers, the traditional and digital industry, the public sector, and academia,” affirms Nikolaos Laoutaris, Research Professor at IMDEA Networks and the Principal Investigator of this transformative project.
The implications of MLEDGE extend far beyond the confines of its immediate objectives. The project will play a pivotal role in the development of Cloud and ML/FL infrastructures in Spain, simultaneously catalyzing national R&D&I efforts. By aligning with the United Nations’ Sustainable Development Goals for 2030, MLEDGE aligns itself with a broader global vision. It actively promotes the sustainable development of efficient networks and FL solutions, empowering practical work that can decisively and positively impact our environment. Noteworthy technological solutions encompassed within MLEDGE’s purview include, but are not limited to:
- Traditional economic domains, such as construction, finance, and healthcare. The project will enable companies to leverage FL to enhance their processes and real-time decision-making through data-driven insights and models.
- Digital economy spheres, particularly in the realm of digital health. Here, MLEDGE will unlock the potential of information derived from mobile devices or wearable technologies, revolutionizing healthcare practices. Additionally, the initiative will contribute to the training of digital advertising models, driving innovation in the advertising landscape.
- Optimization of CloudEdge infrastructures stands as a pivotal functionality within MLEDGE’s overarching framework. Leveraging federated machine learning algorithms, the project aims to achieve remarkable advancements in the field, paving the way for unparalleled efficiency and performance gains.
Conclusion:
MLEDGE’s transformative impact on the market is significant. By integrating federated learning with CloudEdge, the initiative unlocks new possibilities for secure and efficient data processing. This development enables businesses across various sectors to leverage sensitive data and enhance their decision-making processes through machine learning models. MLEDGE’s emphasis on security, efficiency, and sustainability not only drives technological advancements but also fosters economic growth and societal progress. The project’s alignment with the UN’s Sustainable Development Goals and its contribution to national R&D&I further solidify its position as a catalyst for sustainable network solutions in the market.
