BANDMAS：因果启发的语义包调度实现带宽高效多智能体协作
原标题：BANDMAS: Causality-Inspired Semantic Packet Scheduling for Bandwidth-Efficient Multi-Agent Collaboration
AI 摘要
BANDMAS 是一个面向 LLM 多智能体协作的语义包调度框架，通过因果启发的重放估值将消息分解为数据包，仅传输贡献超过资源成本的包，以降低带宽和推理开销。在 SciFact、HotpotQA 和 FanOutQA 数据集上，使用冻结的 Qwen3-4B 模型，BANDMAS 在选定上限下将应用层字节减少 53.2% 至 77.3%，并在所有三个工作负载上取得受约束方法中最高的平均任务指标。
正文节选
Computer Science > Multiagent Systems Title:BANDMAS: Causality-Inspired Semantic Packet Scheduling for Bandwidth-Efficient Multi-Agent Collaboration View PDF HTML (experimental) Abstract:LLM-based multi-agent systems make decisions based on the aggregated information via exchanging messages across specialized agents. Forwarding every generated message among agents increases application-layer traffic. Yet, it introduces tremendous input tokens for agent processing, potentially raising