“…[59][60][61][62] In addition to intriguing works in deep-learning photonics, including the design of nanostructures, [24,25,31] metamaterials [27,29,30] and metasurfaces, [33] holography, [32] our study on active photonic functionalities in disordered platforms will provide additional degrees of freedom in the application of deeplearning to light-matter interactions, paving the way to the handling of spatially-complex and dynamical systems. Although we examined one of the optical responses as an example-angular transmittance-the versatile features of the DNNs will enable engineering of not only the other responses (spectral responses, angular momenta, topology) but also their mixtures: the design of active disorder in the intermediate regime, [9] temporal disorder, [63] and disordered topological phenomena. [7] As shown in the analogy between GST-controlled disorder and target control of complex networks, [3] our machine learning strategy can also be extended to other fields beyond wave mechanics, such as the interpretation and design of evolving complex networks.…”