Abstract:
"In today's rapidly evolving digital landscape, the demand for seamless and personalized user experiences across multiple devices is ever-increasing. Federated Learning (FL) offers a promising solution by enabling collaborative model training across decentralized devices while preserving user data privacy. This project explores a novel framework that leverages FL to synchronize and personalize machine learning models across diverse digital ecosystems, focusing on user behavior rather than merely treating devices as clients.
Central to our approach is the personalization of models based on individual user preferences. By sharing metadata and model updates across user devices, we aim to develop machine learning models that adapt to unique user behaviors, enhancing usability while maintaining robust privacy safeguards by keeping sensitive data decentralized. Our evaluation emphasizes the effectiveness and reliability of this user-as-client approach in real-world scenarios.
Through rigorous testing and benchmarking, we find that our proposed system yields stable results similar to those achieved by FedAvg (McMahan et al., 2023), but with a distinct focus on personalization. This innovative framework not only addresses existing challenges in FL but also opens new avenues for research in the field. Ultimately, our goal is to demonstrate how this approach can foster cohesive and tailored user experiences, paving the way for future advancements in federated learning and personalized technology."