Joint Intelligence Distribution and Fine-Tuning for Multi-Agent Intelligence Manufacturing

The intelligent manufacturing environment imposes extremely high requirements on real-time performance and accuracy. However, due to the dual constraints of communication and computing resources, meeting these requirements poses significant challenges. In this paper, we investigate how to achieve joint optimization of intelligence distribution and fine-tuning in the process of acquiring and applying intelligent models by agents. The framework efficiently provides intelligent models to the agents at a low cost while optimizing the intelligent models to ensure the overall performance of the system. Firstly, to achieve ubiquitous collaboration across computing resources driven by network awareness, we propose a multi-agent intelligent manufacturing architecture based on Computing Power Network (CPN). Secondly, to balance computing and communication resources and enhance model accuracy performance, we formulate a hierarchical optimization framework based on a dual spatial scale approach, combining intelligence distribution and fine-tuning. We further design a joint optimization algorithm based on the Differential Evolutionary (DE) framework and a synergy between Coalition Game Theory (CGT) and Federated Learning (FL) to effectively solve this problem. Finally, extensive simulation experiments validate the effectiveness and superiority of the proposed solution.