SymbXRL: Symbolic Explainable Deep Reinforcement Learning for Mobile Networks

TL;DR

SymbXRL introduces a symbolic explainable reinforcement learning framework for DRL-driven mobile network optimization. By translating DRL states and actions into First-Order Logic terms, it enables interpretable explanations and the Intent-based Action Steering module to steer actions toward operator intents while maintaining or improving performance. The approach combines a Symbolic Representation Generator, an Explanation Engine, and IAS to deliver probabilistic and knowledge-graph analyses of agent behavior, with empirical validation on network slicing and Massive MIMO scheduling tasks showing a median 12% improvement in cumulative reward over baseline DRL and improved training efficiency. This work advances practical deployment of explainable AI in 6G/5G-era networks by offering actionable, rule-based control grounded in symbolic reasoning and demonstrated improvements in both explanation quality and operational performance.

Abstract

The operation of future 6th-generation (6G) mobile networks will increasingly rely on the ability of deep reinforcement learning (DRL) to optimize network decisions in real-time. DRL yields demonstrated efficacy in various resource allocation problems, such as joint decisions on user scheduling and antenna allocation or simultaneous control of computing resources and modulation. However, trained DRL agents are closed-boxes and inherently difficult to explain, which hinders their adoption in production settings. In this paper, we make a step towards removing this critical barrier by presenting SymbXRL, a novel technique for explainable reinforcement learning (XRL) that synthesizes human-interpretable explanations for DRL agents. SymbXRL leverages symbolic AI to produce explanations where key concepts and their relationships are described via intuitive symbols and rules; coupling such a representation with logical reasoning exposes the decision process of DRL agents and offers more comprehensible descriptions of their behaviors compared to existing approaches. We validate SymbXRL in practical network management use cases supported by DRL, proving that it not only improves the semantics of the explanations but also paves the way for explicit agent control: for instance, it enables intent-based programmatic action steering that improves by 12% the median cumulative reward over a pure DRL solution.

Figures (9)

• Figure 1: SymbXRL's architecture and interaction with a drl agent operating on a target environment.
• Figure 2: Probabilistic analysis of SymbXRL showing the probability distribution of decision effects for agent A1's variants under TRF1.
• Figure 3: Correlation density map of input kpi and decisions for two variants of agent \ref{['agent-neu']} favoring embb (first row) or urllc (second row) slices.
• Figure 4: kgs produced by SymbXRL for each implementation of agent \ref{['agent-rice']}.
• Figure 5: Agent \ref{['agent-rice']}: Probability distribution of input KPIs (dtu and mse)