Optimal Resource Management for Wireless Cooperative Edge Learning with Energy Harvesting

This paper proposes a general framework, where energy-harvesting devices may cooperate with one another to perform their learning tasks, in a peer-to-peer fashion. Unlike the classical edge-inference scenarios, where inference tasks are split between an edge device and an edge server (ES), our model considers a peer-to-peer (P2P) wireless network, where each mobile device (MD) can operate both as a client and as a server for other nodes. The framework first establishes node pairs on the basis of the channel state information. Within each pair, nodes can either process their tasks locally or offload them to the associated node, allocating both transmission and computational resources through a Lyapunov optimization procedure. Simulation results, targeted to an image classification task, validate the effectiveness of the proposed algorithm and highlight its potential for broader applications in various scenarios and use cases.