Rapid Design of Metamaterials via Statistical Learning and Deep Learning
报告摘要：Composed of a large number of subwavelength unit cells with designable geometries, metamaterials have been widely studied to achieve extraordinary advantageous and unusual optical properties. However, ordinary computer simulator requires a time-consumingﬁne-tuning toﬁnd a proper design of metamaterial for a speciﬁc optical property, making the design stage a critical bottleneck in large scale applications of metamaterials. In this talk, we investigate the metamaterial design from two different perspectives, which are the framework of computer experiments and deep neural networks, respectively. On one hand, we formulate the multiple related design targets as a multitask design problem. Leveraging on the similarity between different designs, we propose an efﬁcient Bayesian optimization strategy with a parsimonious surrogate model and an integrated acquisition function to design multiple unit cells with very few function evaluations. On the other hand, we propose a Functional Response Conditional Variational Auto-Encoder (FR-CVAE), which takes complex functional response curves as inputs and generates high-quality microstructures whose physical responses match well with the given inputs, to achieve the inverse design of EM metamaterials. Thorough experimental studies demonstrate that the proposed methods are effective and practical for the rapid design of metamaterials.