Ensemble learning: a bidirectional framework for designing data-driven THz composite metamaterials

Metamaterials present revolutionary routes to manipulate the behavior of electromagnetic waves. The well-designed metamaterial can exhibit exotic functionalities, such as perfect absorption, holography, beam steering, optical nonlinear generation, and various functional interfaces. However, those designs currently rely on trial-and-error and case-by-case numerical simulations to achieve target responses, which usually requires huge computing resources and expertise related to metamaterials. In this study, we propose a machine-learning-assisted bidirectional ensemble learning framework for designing composite metamaterial absorbers at 0.3–2.0 THz. The proposed framework is a guide to reveal the intricate and nonintuitive relationship between a composite metamaterial structure and its absorption spectrum from previously known datasets, which circumvents the limitation of numerical simulation. This framework not only effectively realizes the forward prediction of the absorption spectrum, but also can retrieve composite metamaterial structure parameters from a given spectrum.