Prioritized Value-Decomposition Network for Explainable AI-Enabled Network Slicing

2025-01-27

AI Summary

To address the explainability gap in AI-driven resource allocation for 6G network slicing—stemming from the “black-box” nature of conventional AI models—this paper proposes a multi-agent reinforcement learning (MARL) framework integrated with eXplainable Artificial Intelligence (XAI). The core contribution is a novel priority-driven value decomposition mechanism: it ranks slice contributions via attention-weighted ordering, incorporates SHAP-inspired causal attribution, and embeds an end-to-end differentiable explanation module. This enables transparent, traceable decision-making and explicit visualization of the compute-performance trade-off. Experiments demonstrate that, compared to independent learning and standard Value Decomposition Networks (VDN), the proposed method achieves 67% and 16% higher throughput, respectively, and reduces latency by 35% and 22%. Consequently, network operators gain enhanced understanding, trust, and controllable intervention capability over slicing policies.

Technology Category

• Humans and AI: Explainable AI (XAI) for Human Understanding
• Philosophy and Ethics of AI: Accountability, Interpretability & Explainability
• Multiagent Systems: Adversarial Agents

Application Category

• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Network slicing aims to enhance flexibility and efficiency in next-generation wireless networks by allocating the right resources to meet the diverse requirements of various applications. Managing these slices with machine learning (ML) algorithms has emerged as a promising approach however explainability has been a challenge. To this end, several Explainable Artificial Intelligence (XAI) frameworks have been proposed to address the opacity in decision-making in many ML methods. In this paper, we propose a Prioritized Value-Decomposition Network (PVDN) as an XAI-driven approach for resource allocation in a multi-agent network slicing system. The PVDN method decomposes the global value function into individual contributions and prioritizes slice outputs, providing an explanation of how resource allocation decisions impact system performance. By incorporating XAI, PVDN offers valuable insights into the decision-making process, enabling network operators to better understand, trust, and optimize slice management strategies. Through simulations, we demonstrate the effectiveness of the PVDN approach with improving the throughput by 67% and 16%, while reducing latency by 35% and 22%, compared to independent and VDN-based resource allocation methods.

Problem

• Machine Learning
• Network Slicing
• Transparency in AI

Innovation

• PVDN
• Resource Allocation Transparency
• Performance Enhancement