Deep Reinforcement Learning (DRL) is a promising tool for adaptive network control, but most DRL agents remain reactive: they make decisions from past and current measurements and cannot fully exploit short-term forecasts of changing network conditions. Forecast-aware agents can act proactively, yet their decisions are difficult to understand, which makes operators unsure whether predictions are actually useful or worth the added complexity. This talk presents SIA, a symbolic interpretability framework for anticipatory DRL in network control. SIA converts network KPIs and agent actions into human-readable symbolic representations, organizes them through per-KPI knowledge graphs, and introduces an Influence Score to explain how current observations and future predictions shape each decision in real time. Across adaptive video streaming, massive MIMO scheduling, and RAN slicing, SIA reveals hidden agent behaviors, guides better agent design, and improves performance without costly retraining.

About MohammadErfan Jabbari

MohammadErfan Jabbari is a Research Engineer at IMDEA Networks Institute, where he works with the NDS and RAINET research groups. He is currently pursuing a Master’s degree in Machine Learning for Health at Universidad Carlos III de Madrid. His research focuses on deep reinforcement learning, network control, and interpretable AI methods for intelligent network systems.

This event will be conducted in English