BANDMAS: Causality-Inspired Semantic Packet Scheduling for Bandwidth-Efficient Multi-Agent Collaboration

2026-08-01

AI Summary

This work addresses the challenge of communication redundancy and inference latency in multi-agent systems, where full-message transmission often compromises efficiency without proportionally improving decision quality. To reconcile this trade-off, the authors propose a causality-inspired, semantic packet-level scheduling mechanism. Messages are decomposed into semantically meaningful packets—such as evidence or request types—and their predictive contributions are evaluated via causal replay. Transmission occurs only when the expected utility exceeds the associated resource cost, with scheduling dynamically adapted under constraints including bandwidth, latency, deadlines, and receiver context. Experiments on SciFact, HotpotQA, and FanOutQA using a frozen Qwen3-4B model demonstrate that the method reduces application-layer communication by 53.2%–77.3% while maintaining the highest average task performance, substantially outperforming existing constrained optimization approaches.

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 inference latency and computational overhead. Existing approaches attempt to address the above issues by pruning agents or discarding redundant messages. Nevertheless, such agent-level or message-level optimization results in insufficient evidence supporting for final decisions or still containing redundant message transmissions. To address these challenges, we propose BANDMAS, a multi-agent collaboration framework that models inter-agent communications as task-oriented traffic, which enables efficient transmission via causality-inspired replay valuation. Specifically, we decompose messages into several data packets by analyzing their semantic features such as evidence and requests. The system only transmits these packets if their predicted replay-derived contribution exceeds their resource cost. Consequently, BANDMAS is able to adaptively schedule communication packets while adhering to bandwidth, latency, deadline, and receiver context constraints. On frozen Qwen3-4B traffic across SciFact, HotpotQA, and FanOutQA, our framework reduces application-layer bytes by 53.2\% to 77.3\% at selected caps and attains the highest mean task metric among constrained methods on all three workloads.

Problem

• multi-agent collaboration
• bandwidth efficiency
• message redundancy
• communication overhead
• semantic packet scheduling

Innovation

• causality-inspired scheduling
• semantic packet decomposition
• bandwidth-efficient multi-agent collaboration
• replay valuation
• LLM-based multi-agent systems