Peer-to-peer federated learning

Abstract

This Final Year Project (FYP) explores the integration of Peer-to-Peer Federated Learning (P2P FL) on Android devices, leveraging TensorFlow Lite alongside Wi-Fi Direct and Bluetooth for decentralized machine learning (ML) model training directly on mobile devices. The study aims to harness the collective computational power of smartphones to collaboratively improve ML models while ensuring user data privacy. Experiments conducted across various Android devices, demonstrate the framework’s adaptability to different hardware and software environments, ensuring efficient model training and synchronization despite device diversity. The project highlights the potential of P2P FL in making artificial intelligence more accessible and customizable across a broad spectrum of devices. However, challenges such as communication overhead, scalability, and mobile device limitations are acknowledged, emphasizing the need for ongoing research and optimization in P2P FL methodologies. This work contributes to the evolving field of FL, offering insights into its application on mobile platforms and outlining future directions for leveraging mobile devices in collaborative learning scenarios.