Federated Learning for Personalized Recommendation

Abstract—User experiences are improved by personalized recommendation systems, which have various uses in e-commerce, entertainment, health, etc. But there are privacy and legal problems with the centralized data collection. As an alternative to centralized data collection, Federated Learning (FL) offers model training on user devices. It allows model training without gaining access to personal information. The purpose of this seminar is to showcase FL's potential in personalized recommendations for shopping, healthcare, and content delivery. Prescriptive analytics combined with federated learning can help pet owners adopt pets while protecting their privacy, as demonstrated by a PETNOVA case study. There are issues with algorithm performance, fairness, and model heterogeneity that will be covered. Also go over the future developments of hybrid learning, secure aggregation, and differential privacy.