Physics-inspired framework proposed for interpretable attribution in cyber-physical IoT systems

A new paper proposes a physics-inspired framework to address interpretability challenges in large-scale, hybrid cyber-physical systems, including industrial IoT environments. The method models variable dependencies through an undirected, energy-based representation derived from statistical mechanics, enabling dependency-aware attribution without explicitly recovering directed causal graphs.

The authors argue that traditional explainability methods, which focus on correlations, are insufficient for interventional questions in high-risk domains. Their approach analyzes how variations in the energy landscape reflect the influence of individual components, supporting reasoning about perturbation effects across hybrid interactions and abnormal behavior detection.

Empirical evaluation on an industrial IoT testbed with hybrid continuous and discrete variables reports higher attribution accuracy, improved robustness, and better scalability compared to state-of-the-art graph-based approaches. The framework is positioned as a complementary tool for human interpretation and downstream predictive and diagnostic tasks, rather than a full recovery of generative dynamics.

While demonstrated in industrial IoT security, the authors suggest the method generalizes to other high-dimensional cyber-physical and socio-technical systems that require principled, structural explanations.