Executive Summary : | The demand for efficient machine learning models in real-time Internet of Things applications is growing due to the need for high-performance models for automation, monitoring, and data analytics. However, the federated learning paradigm, a decentralized technology, faces challenges due to the scarcity of high-quality data from sensitive sources like industrial data silos and electronic health records. This project aims to build upon the existing federated learning framework by proposing a robust and collaborative fog-aided federated learning framework for enhanced resiliency of real-time applications. The project plans to replace the single central server with fog nodes to increase model reliability and avoid destruction due to single point failure. The proposed framework will also be optimized by incorporating an incentive mechanism for collaborative model training and Deep Reinforcement Learning algorithms to improve model efficiency. The Stackelberg game theory will be used to motivate fog/edge nodes to participate in the collaborative training process, addressing the issue of data inconsistency and reducing false predictions. The proposed model will be empirically demonstrated by prototyping it using a real-time application like water irrigation control in rural areas near Lucknow City. |