Title: Machine learning identifies scale-free properties in disordered materials

Abstract: The vast amount of design freedom in disordered systems expands the parameter space for signal processing, allowing for unique signal flows that are distinguished from those in regular systems. However, this large degree of freedom has hindered the deterministic design of disordered systems for target functionalities. Here, we employ a machine learning (ML) approach for predicting and designing wave-matter interactions in disordered structures, thereby identifying scale-free properties for waves. To abstract and map the features of wave behaviours and disordered structures, we develop disorder-to-localization and localization-to-disorder convolutional neural networks (CNNs). Each CNN enables the instantaneous prediction of wave localization in disordered structures and the instantaneous generation of disordered structures from given localizations. We demonstrate that CNN-generated disordered structures have scale-free properties with heavy tails and hub atoms, which exhibit an increase of multiple orders of magnitude in robustness to accidental defects, such as material or structural imperfection. Our results verify the critical role of ML network structures in determining ML-generated real-space structures, which can be used in the design of defect-immune and efficiently tunable devices.