Fig. 5: Inverse design approach based on sub-manifold learning and neural adjoint method.
From: Machine learning for knowledge acquisition and accelerated inverse-design for non-Hermitian systems
a Convex-hulls of the feasible regions for non-Hermitian structures for the training transmission data in latent space. b–d Desired spectral responses (using Transfer Matrix Method: TMM and Machine Learning: ML) versus normalized frequency ωa/2πc, with a the dimension of the scatterer and c the speed of light in free space, designed with adaptive gradient descent method. The initial seed is obtained within feasible sub-manifolds of lossy, gain, and balanced non-Hermitian systems.
