Cooperative UAV swarms for zero knowledge verification of edge generative AI using trust-aware multiagent learning

Area of Science:

• Computer Science
• Artificial Intelligence
• Cybersecurity

Background:

• Generative AI integration in edge computing raises concerns about Model-as-a-Service integrity.
• Edge servers may not adhere to generative models to reduce costs, necessitating validation without compromising proprietary information.

Purpose of the Study:

• To propose a secure and privacy-preserving framework for verifying edge-based generative AI inference.
• To address the challenge of validating AI models in edge environments while protecting sensitive data.

Main Methods:

• A cooperative unmanned aerial vehicle (UAV)-swarm-enabled zero-knowledge verification framework.
• Edge servers generate interactive cryptographic zero-knowledge proofs for AI execution verification.
• UAV swarms perform verification operations under mobility and energy constraints.
• A trust-based multi-agent reinforcement learning approach for UAV swarm behavior planning.

Main Results:

• The proposed framework significantly improves verification timeliness and reduces malicious server detection delay.
• Demonstrates enhanced energy efficiency and scalability compared to baseline schemes.
• The 'age of verification' metric effectively prioritizes risky edge servers.

Conclusions:

• Integrating cooperative UAV swarms, trust-aware verification, and multi-agent learning provides reliable generative AI services in edge computing.
• The framework ensures secure, privacy-preserving verification of AI inference at the edge.