Finding the Weakest Link: Adversarial Attack against Multi-Agent Communications

Maxwell Standen, Junae Kim, Claudia Szabo

Keywords: adversarial attack, adversarial machine learning, multi-agent reinforcement learning, multi-agent communications, robustness, security

Abstract: Multi-agent systems rely on communication for information sharing and action coordination. However, this communication is vulnerable to attacks, and a system is only as robust as its weakest link. To understand the weakest links in a Multi-Agent Reinforcement Learning-trained system, we investigate single-victim communication perturbation attacks and propose methods that use gradient information from the Jacobian to identify which messages, agent, and timesteps are most susceptible to attack and have the greatest impact on the system. We enhance these methods with two proposed adversarial loss functions that trade-off attack success for attack impact. These loss functions help identify the weakest links and create more effective perturbations. We empirically demonstrate the effectiveness of our methods against two different multi-agent communication methods in navigation, PredatorPrey, and TrafficJunction environments. Our results show that our novel message selection method achieves a similar or greater impact than random message selection across almost all tested scenarios. Our victim selection, tempo, and loss functions improve attack effectiveness in eight scenarios and similar performance to previous attacks in nine scenarios out of the thirty tested scenarios.

Area: Learning and Adaptation (LEARN)

Generative A I: I acknowledge that I have read and will follow this policy.

Submission Number: 1148