REDEEM Project at “Les Rencontres de l’IA Embarquée” – March 7, 2025

On March 7, 2025, the REDEEM project participated in Les Rencontres de l’IA Embarquée, hosted at Bercy. Cedric, coordinator of the REDEEM project, co-hosted a matchmaking session titled “Distributed and Federated AI: From Edge to Cloud” alongside Hugo Miralles from MANTA (Inria Startup Studio).

Secure Federated Graph-Filtering for Recommender Systems

Recommender systems rely on graph-based filters, but their centralized computation poses privacy and security risks. Our work introduces two decentralized frameworks that securely compute key graph components without centralizing user data. Leveraging Multi-Party Computation and distributed spectral methods, we ensure privacy while maintaining predictive accuracy. Our second approach further optimizes efficiency through low-rank approximations, balancing performance and communication costs. Experiments on benchmark datasets show that our methods match state-of-the-art accuracy while preserving data confidentiality. This work paves the way for privacy-preserving, decentralized recommender architectures.