XAI-Guided Conservative Decentralized Execution for Offline Multi-Agent Network Slicing

Offline learning reduces deployment risks: X-CODE operates without requiring real-time environmental interaction or communication between agents during deployment, making it safer and more practical to implement in critical network infrastructure compared to traditional online reinforcement learning approaches.

Explainability improves multi-agent coordination: The paper demonstrates that incorporating explainable AI (XAI) principles into reward shaping helps agents make better independent decisions during decentralized execution, achieving "zero observed resource-conflict events" in testing while minimizing delays—suggesting that understanding why agents make decisions improves their ability to coordinate without direct communication.

Addresses a real infrastructure challenge: Network slicing requires allocating limited resources among competing services (like video streaming, medical devices, and autonomous vehicles) that need different performance levels; this work tackles that practical problem by enabling intelligent, autonomous resource management without constant human oversight or inter-system communication.

No communication overhead in deployment: Unlike many multi-agent systems that require agents to continuously exchange information, X-CODE allows agents to operate independently after training, reducing communication costs and system complexity in real-world 6G network implementations.