Security-Enhancing Mechanisms to Strengthen Privacy on Federated Learning Based Recommendation Systems
João Gaspar, José Pessoa, Bruno Ribeiro, Diogo Martinho, Joaquim Santos, Goreti Marreiros · 2025
In the current landscape of Artificial Intelligence solutions, Federated Learning emerges as one of the most promising techniques due to its collaborative nature across distinct clients as well as a privacy-preserving technique in information exchange. This preservation is accomplished by the node contributions being made in the form of model updates instead of sharing the raw data with the global server. However, there are challenges in this field that remain, such as inference attacks or model manipulations. This paper provides an overview regarding existing concerns on centralized data and how the concept of federated may be a valid tool to mitigate centralization problems, with particular focus in recommendation systems. The study then addresses different approaches to increase the privacy guarantees of federated learning, with resource to additional privacy mechanisms.