Comprehensive survey on communication structure adaptive control and collaborative optimization for multiagent systems based on deepseek large language model
Bangyin Xiang · 2025
This survey provides a comprehensive overview of adaptive communication topology control and collaborative optimization in multi-agent systems (MAS), with a specific emphasis on the capabilities of the DeepSeek large language model. We first review mainstream communication structure regulation methods, such as reinforcement learning–based topology evolution and gradient-driven graph reconfiguration. The survey then investigates how DeepSeek enables endto-end coordination by translating natural language instructions into structural adaptations and multi-agent policies. Its strengths in semantic parsing, instruction alignment, and multi-objective optimization generation make it a compelling paradigm for MAS cooperation under complex environments. We highlight the generalization and transferability of DeepSeek across diverse decision-making scenarios and discuss its role in relaxing traditional modeling constraints. Finally, we outline open challenges related to real-time reasoning, interpretability, and adversarial robustness. This work aims to provide both theoretical guidance and practical insights for LLM-empowered MAS coordination frameworks.