Advancing photonic design with topological latent diffusion generative model
Yuheng Chen, Michael Bezick, Blake Wilson, Ömer Yeşilyurt, Alexander V. Kildishev, Alexandra E. Boltasseva, Vladimir M. Shalaev · 2024
Conventional photonic design often relies on inefficient trial-and-error. The proposed Topological Latent Diffusion Model (TLDM) captures high-level features from topology dataset and outperforms state-of-the-art Generative Adversarial Networks and Variational Autoencoders methods in high-efficiency metasurface design.