Collaborative Offloading of AI Workloads in Heterogenous IoT

The proliferation of constrained devices in Edge networks presents a unique opportunity for executing complex AI tasks despite their individual limitations or the task complexity itself. IoT Edge devices are often underutilized, as they are inherently optimized for predefined tasks and bandwidths to maintain cost-effectiveness and reduce complexity. However, constrained devices, within both the immediate and broader network, can collaborate by sharing their processing power and resources, effectively forming a distributed computing network. By pooling their resources, they can collectively tackle complex AI tasks that would be impossible for any single device to handle alone. This collaborative approach also enhances their flexibility in scaling to meet unpredictable operational demands, enabling the deployment of more demanding AI services. This paper proposes a multi-tiered framework for AI task management by leveraging the resources of the broader network through a collaborative network of coordinators and AI task offloading. This framework serves as a steppingstone toward systems that can efficiently marshal network resources to address AI workloads. Furthermore, by leveraging the visibility of numerous nodes across diverse environments, this framework can facilitate an Edge-centric approach to federated learning. The decentralized learning approach empowers constrained edge devices to collaboratively train and improve their AI models based on the unique data encountered in their deployment environments. This framework not only results in more accurate and robust AI models but also enables more adaptive applications post-deployment.