A Distributed Topology-Protecting Collaboration Algorithm: Design and Performance Analysis.

Area of Science:

• Computer Science
• Control Theory
• Cybersecurity

Background:

• Multiagent systems (MASs) rely on interaction topology for collaboration.
• Topology inference techniques, while useful, pose privacy risks from attackers.
• Existing topology protection methods are insufficient against sophisticated inference attacks.

Purpose of the Study:

• To propose a distributed algorithm for multiagent systems to defend against topology inference attacks.
• To enhance the privacy and security of MASs by protecting their interaction topology.
• To address the unresolved challenges in topology protection for MASs.

Main Methods:

• Developed a novel distributed collaboration algorithm for MASs.
• Introduced a new noise term to increase agent state irregularity without compromising convergence.
• Implemented a weight-selecting strategy to degrade topology inference accuracy by selecting subtopologies.

Main Results:

• The proposed algorithm effectively increases state irregularity in a distributed manner.
• The weight-selecting strategy significantly degrades topology inference accuracy.
• Theoretical analysis derived mean-square convergence factors and nonasymptotic error bounds.
• Extensive simulations validated the algorithm's effectiveness in topology protection.

Conclusions:

• The novel distributed algorithm provides a robust defense against topology inference attacks in MASs.
• The approach enhances system privacy by obscuring interaction topology.
• This work offers a significant advancement in securing MASs against adversarial inference.