SIA: Symbolic Interpretability for Anticipatory Deep Reinforcement Learning in Network Control

TL;DR

SIA addresses the opacity of forecast-aware deep reinforcement learning for network control by introducing Symbolic Interpretability for Anticipatory DRL. It combines Symbolic AI with per-KPI Knowledge Graphs to produce real-time, global and local explanations, anchored by a novel Influence Score that disentangles current-state and forecast effects. The framework includes an optional Action Refinement module that improves performance without retraining, demonstrated across ABR, Massive MIMO scheduling, and RAN slicing with sub-millisecond latency and notable gains (e.g., up to 9% bitrate increase and 25% reward boost). By making anticipatory control transparent and tunable, SIA lowers barriers to deployment of proactive network management in 5G/6G contexts.

Abstract

Deep reinforcement learning (DRL) promises adaptive control for future mobile networks but conventional agents remain reactive: they act on past and current measurements and cannot leverage short-term forecasts of exogenous KPIs such as bandwidth. Augmenting agents with predictions can overcome this temporal myopia, yet uptake in networking is scarce because forecast-aware agents act as closed-boxes; operators cannot tell whether predictions guide decisions or justify the added complexity. We propose SIA, the first interpreter that exposes in real time how forecast-augmented DRL agents operate. SIA fuses Symbolic AI abstractions with per-KPI Knowledge Graphs to produce explanations, and includes a new Influence Score metric. SIA achieves sub-millisecond speed, over 200x faster than existing XAI methods. We evaluate SIA on three diverse networking use cases, uncovering hidden issues, including temporal misalignment in forecast integration and reward-design biases that trigger counter-productive policies. These insights enable targeted fixes: a redesigned agent achieves a 9% higher average bitrate in video streaming, and SIA's online Action-Refinement module improves RAN-slicing reward by 25% without retraining. By making anticipatory DRL transparent and tunable, SIA lowers the barrier to proactive control in next-generation mobile networks.

Figures (11)

• Figure 1: Agents using future network bandwidth estimates achieve higher qoe by acting proactively.
• Figure 2: Architecture of SIA, showing the core modules and information flow from raw kpi to the explanations
• Figure 3: Symbolic fol representations for the kpi of our evaluation agents.
• Figure 4: Policy graph of the reactive agent (A1-R).
• Figure 5: Policy graph of the proactive agent (A1-P).