Trustworthy Al in Mobile Radio Networks:Explainability, Causality, Uncertainty Quantification

Abstract

Mobile radio networks beyond 5G become increasingly complex, making the role of machine learning ever more significant, and the interpretability and reliability of model decisions critical. First, I examine how explainable modeling (XAl)—for example, additive local feature attribution methods such as SHAP-can reveal causal relationships between network configuration and performance indicators (KP|s). We introduce new attribution techniques that are better aligned with causal dependencies, thereby improving interpretability [Kelen et al, ICDM'25]. Motivated by radio network forecasting, we then address the challenge of uncertainty quantification in regression problems. We propose a non-deterministic neural network regression framework optimized using a sample-based approximation of the Continuous Ranked Probability Score (CRPS). This enables distribution-free learning of aleatoric uncertainty, providing well-calibrated probabilistic predictions [Kelen et al, ICLR'25].

Finally, we discuss post hoc, nonparametric recalibration methods inspired by new statistical tests for assessing model calibration, in order to enable reliable decision-making in complex, high-risk network environments.

Speaker Biography

András Benczúr received his Ph.D. in Applied Mathematics at the Massachusetts Institute of Technology in 1997.

Andras is a Hungarian computer scientist and research leader serving as the Scientific Coordinator of Hungary's Artificial Intelligence National Laboratory and as Head of the Al Laboratory at the Institute for Computer Science and Control of the Hungarian Research Network. His work focuses on the development and application of data-driven and machine learning methods, with particular emphasis on networks, large-scale data analysis, and trustworthy Al systems in healthcare, telecommunications, and manufacturing. In his leadership roles including Data Industry Working Group leader of the Artificial Intelligence Coalition Hungary, he coordinates national research efforts, fosters collaboration between academia and industry, and contributes to shaping Hungary's strategic direction in artificial intelligence research and innovation.

Light refreshments will be served after the colloquium.