Adversarial Autoencoders for Metasurface Design Optimization (invited)

We have coupled adversarial autoencoder deep generative network with adjoined topology optimization technique for metasurface design optimization. We have demonstrated that such approach ensures not only efficient optimization search of highly efficient metasurface designs but also provides unparalleled control over compressed design space distribution. The latter fact assures scalability of the approach to highly-constrained optimization problems.