From Graphs to Gradients: Physics-Inspired Structural Attribution for Cyber-Physical IoT Systems and Beyond

Short summary

This paper introduces a physics-inspired framework using energy-based representations to perform structural attribution in cyber-physical IoT systems, bypassing the impractical recovery of directed causal graphs. By modeling variable dependencies through an undirected energy landscape, the approach enables dependency-aware attribution and perturbation reasoning across hybrid continuous and discrete interactions. Empirical simulations on an industrial IoT testbed show higher attribution accuracy, robustness, and scalability compared to state-of-the-art graph-based methods.

• •Proposes an energy-based, undirected framework for structural attribution in cyber-physical IoT systems, avoiding impractical directed causal graph recovery
• •Analyzes energy landscape variations to attribute influence to individual components and reason about perturbation effects across hybrid interactions
• •Empirically demonstrates superior accuracy, robustness, and scalability over graph-based approaches on an industrial IoT testbed

Generated with AI, which can make mistakes.