Empowering Dynamic Scheduling in IoT Networks Through Collaborative Policy Learning via Federated Reinforcement Techniques
M. Thirunavukkarasu, S. Thaiyalnayaki, Naduvathezhath Nessariose Jose, S. Maheshwari, B. S. Pavithra, G. Malathi · 2024
In the burgeoning landscape of Internet of Things (IoT) networks, efficient management of resources is paramount for ensuring optimal performance and resource utilization. Dynamic scheduling, particularly in the context of cloud-edge-terminal IoT networks, presents a significant challenge due to the diverse and dynamic nature of connected devices and their varying computational requirements. Traditional centralized approaches to scheduling may prove inadequate in such dynamic environments, necessitating the exploration of novel techniques. This project proposes a pioneering approach to address the dynamic scheduling challenges in IoT networks by leveraging collaborative policy learning through federated reinforcement techniques. The proposed framework harnesses the power of federated learning, a decentralized machine learning paradigm, to collectively train policies for dynamic scheduling tasks across distributed edge and terminal devices while preserving data privacy and security. Key components of the proposed framework include a collaborative learning architecture that orchestrates the exchange of policy updates among edge and terminal devices, enabling them to adaptively refine their scheduling policies based on local observations and feedback. Reinforcement learning serves as the underlying mechanism for policy optimization, allowing devices to learn and adapt to evolving network conditions and user demands over time. By decentralizing the learning process and leveraging the collective intelligence of edge and terminal devices, the proposed framework offers several advantages. These include improved scalability, reduced communication overhead, and enhanced resilience to network failures. Furthermore, the federated approach ensures data privacy and regulatory compliance by keeping sensitive information localized to individual devices. To evaluate the effectiveness of the proposed framework, comprehensive simulations and real-world experiments will be conducted using representative IoT network scenarios. Performance metrics such as throughput, latency, and energy efficiency will be measured to assess the efficacy of the collaborative policy learning approach compared to traditional centralized scheduling techniques. Overall, this project aims to advance the state-of-the-art in dynamic scheduling for IoT networks by harnessing the potential of collaborative policy learning via federated reinforcement techniques, thereby paving the way for more efficient and scalable resource management in future IoT deployments.