This document discusses deep learning inverse methods for artificial electromagnetic material design. It reviews state-of-the-art approaches and presents a survey of deep learning inverse techniques. Benchmark problems are proposed to efficiently determine which deep learning technique is best suited to solving different design challenges. As problems become more ill-posed, the neural adjoint with boundary loss generates better solutions faster, regardless of simulation constraints. On simpler problems, direct neural networks perform better with limited simulations, while mixture density networks and conditional variational auto-encoders can improve with continued sampling. The document calls for more benchmarking and model/data sharing to better evaluate progress in this field.

Nverse Deep Learning Methods and Benchmarks For Artificial Electromagnetic Material Design PDF

The list of books you might like