In the field of Artificial Intelligence, interpretable explanation methods aim to uncover underlying causes and their effects, enabling a deeper understanding of system behavior under varying inputs. Unlike traditional explainability methods that mainly highlight correlations between input and output variables, causal explanations focus on interventional questions. This approach provides more robust insights, assisting users in understanding automated decisions, particularly in high-risk domains.
However, recovering an explicit directed causal structure in large-scale, hybrid cyber-physical systems with feedback loops and partial observability is often impractical. This paper presents a novel framework inspired by statistical mechanics, modeling variable dependencies through an undirected, energy-based representation of cyber-physical IoT systems.
Our approach enables rigorous dependency-aware attribution by analyzing how variations in the energy landscape reflect the influence of individual components, without the need to recover a directed causal graph.
It also supports reasoning about perturbation effects across hybrid interactions, providing reliable explanations for abnormal behaviors. We empirically examined our framework through simulations on an industrial IoT testbed with hybrid continuous and discrete variables, demonstrating higher attribution accuracy, improved robustness, and better scalability than state-of-the-art graph-based approaches.
While the attributions are not intended to fully recover the system's generative dynamics, they provide valuable, dependency-aware explanations supporting both human interpretation and downstream predictive and diagnostic tasks. Although demonstrated in industrial IoT security, our framework also applies to other high-dimensional cyber-physical and socio-technical systems requiring principled, structural explanations.
Blogger's Review: The energy-based attribution framework proposed in this paper offers a novel perspective for effectively analyzing variable dependencies in complex cyber-physical systems. Its advantages in accuracy and scalability are noteworthy, especially in high-risk applications. This method allows researchers to better understand system behavior and enhances predictive and diagnostic capabilities.