Probability-Density-Based Deep Learning Paradigm for the Fuzzy Design of Functional Metastructures
Example I of the PDN-based inverse design and experimental demonstration. (a) The output mixture Gaussian for a target transmission spectrum, which is visualized in a 2D plot by reducing the data dimensions from 5 to 2 via the principal component analysis. (b) Exact positions of the local maxima visualized in the contour diagram, where each maximum is mapped to a metastructure. The local maxima and the corresponding metastructures are labelled as ,, and , respectively. (c) The comparison among the target transmission spectra, the PDN predicted, and the experimentally measured ones for the three different metastructures ,, and from the top down. Each predicted ,, and structure has different radii per layer, but all get very similar desired transmissions to the target ones. The radii of cylindrical air channels in the labelled structure and the predicted structures are appended on the right side for reference.