Successive Training of a Generative Adversarial Network for the Design of an Optical Cloak

TL;DR

This paper introduces a deep learning-based optimization method using a DCGAN with a feedback loop to design 2D optical cloaks, improving the prediction and discovery of optimal cloaking geometries.

Contribution

It presents a novel iterative training approach for a DCGAN to effectively design optical cloaks with improved accuracy.

Findings

Successful design of 2D optical cloaks using the proposed method

Enhanced prediction of optimal geometries through iterative retraining

Demonstrated effectiveness of feedback loop in GAN training

Abstract

We present an optimization algorithm based on a deep convolution generative adversarial network (DCGAN) to design a 2-Dimensional optical cloak. The optical cloak consists in a shell of uniform and isotropical dielectric material, and the cloaking is achieved via the geometry of the shell. We use a feedback loop from the solutions of the DCGAN to successively retrain it and improve its ability to predict and find optimal geometries.