Attention-based Hierarchical Variational Information Bottleneck for Robust Multi-Agent Communication under Variable Bandwidth

2026-09-28

AI Summary

This study addresses the challenges of prefix truncation and transmission robustness in multi-agent communication under bandwidth constraints by proposing the AH-VIB model. This method integrates a variational information bottleneck (VIB) with an autoregressive sequence generation mechanism, incorporating a hierarchical robust loss function optimized end-to-end via attention mechanisms and reinforcement learning. The core innovation lies in achieving graceful degradation under bandwidth limitations: even when messages are partially truncated, the effective transmission of critical information is preserved. Experimental results demonstrate that under the most stringent bandwidth conditions in cooperative tasks, the proposed model significantly improves communication reliability and average returns. Consequently, this work provides an efficient solution for multi-agent coordination operating within limited bandwidth environments.

Abstract

Learning-based multi-agent communication under limited bandwidth does not only require deciding what to communicate, but also structuring messages so that partial transmissions remain useful. We study this problem under prefix truncation, where only the first part of each message is received. To address it, we propose \textbf{AH-VIB}, an attention-based autoregressive variational communication model that combines a variational information bottleneck (VIB) with sequential message generation and a hierarchical robustness loss. We evaluate AH-VIB on a custom cooperative object-inspection and occupancy-mapping task, where agents equipped with a limited field-of-view sensor coordinate to scan inspection objects in an occupancy-grid world, under variable and fixed bandwidth conditions, and compare it against MADDPG, CommNet, a flat VIB baseline, and an autoregressive MLP ablation. AH-VIB achieves competitive mean return while improving performance reliability under the most constrained bandwidth conditions. These results indicate that AH-VIB improves the reliability and graceful degradation of learned communication under bandwidth constraints.

Problem

• Multi-Agent Communication
• Variable Bandwidth
• Prefix Truncation
• Information Bottleneck
• Graceful Degradation

Innovation

• Variational Information Bottleneck
• Multi-Agent Communication
• Autoregressive Message Generation
• Hierarchical Robustness Loss
• Graceful Degradation