A Token/KV-Cache Communication Media Selection and Resource Allocation Strategy for Multi-Agent Collaboration
Researchers have developed a new strategy for optimizing communication in multi-agent collaboration systems, particularly those integrating large language models with 6G networks. The proposed method jointly selects communication media and allocates wireless resources to minimize end-to-end latency. It demonstrates that neither token-based nor KV-cache-based transmission is universally superior, with optimal performance depending on factors like computational resources and channel conditions. The developed algorithm, JMSRA, adaptively coordinates interaction media and bandwidth to achieve significantly reduced latency compared to existing baselines. AI
IMPACT Optimizes communication for multi-agent AI systems, potentially improving efficiency in future networked AI applications.