Fig. 2.

Experimental verification of the Neuroute embedded with the istGAN. (A) A simplified flowchart for generating the Neuroute comprises 2 primary components: the construction of the istGAN and experimental validation. (B) The istGAN architecture schematic. Considering the one-to-many mapping between the scattering matrix and the on-demand wavefront, we divide istGAN into forward and inverse models, and the convergence condition depends on cyclic iterations of the 2 networks. The istGAN is specially designed for non-invasive self-adaptive information states’ acquisition inside scattering spaces, drawing on the principle of the cGAN and the tandem neural network. (C) Experimental setup. (D) The single-target Neuroute for information states’ acquisition is simulated by utilizing a sole measurement of the scattering matrix. See Movie S1 for detailed information about processing. (E) Corresponding experimental results using equipment in (C).