Toward Efficient Collaborative Inference in IIoT: A Diffusion Model and DRL-Based Approach

Deep Neural Networks (DNNs) have been extensively applied in the Industrial Internet of Things (IIoT) to support intelligent manufacturing and industrial automation. However, the limited computing and energy resources of IIoT devices hinder the efficiency of on-device DNN inference due to increased energy consumption and latency. To mitigate these limitations, collaborative inference and adaptive sampling techniques have emerged as effective approaches to dynamically allocate computational workloads between IIoT devices and edge servers, thereby reducing on-device burdens and improving overall system performance. In this paper, we propose an edge-assisted IIoT system that enables collaborative inference through a novel Diffusion Model-based Deep Reinforcement Learning (Diffusion-DRL) algorithm. The algorithm jointly optimizes sampling and task offloading strategy to minimize terminal energy consumption while ensuring inference accuracy and meeting latency requirements under complex constraints. By combining the generative power of the diffusion models with the decision-making capabilities of deep reinforcement learning, Diffusion-DRL effectively handles multi-dimensional discrete action spaces. Experimental results demonstrate that the proposed approach significantly reduces latency and energy consumption, while enhancing inference accuracy, offering an effective solution for intelligent collaborative inference in IIoT systems.