Optimizing Task Allocation for DNN Inference on Edge Devices

As artificial intelligence continues to evolve, the need to deploy models at the network edge becomes increasingly critical. A key challenge in such environments is the effective allocation of tasks across devices. To tackle this challenge, we formulate the execution of DNN inference tasks on edge devices as an optimization problem and design a one-to-one task allocation scheme that optimizes the total execution time for a given set of tasks. Our method is both device-and task-aware, employing a greedy algorithm to create task-device mappings. We demonstrate the viability of this approach through comparisons with random allocation and nearest-device strategies, showing that our scheme consistently outperforms these alternatives.