Inverse Design of Magnetic Fields Using Deep Learning

We here present a novel deep learning (DL) approach for designing structures of permanent magnets. The challenge the DL method in this kind problem is to learn mapping from desired magnetic field simple structure, i.e., an inverse design approach. demonstrate by training six different standard convolutional neural network (CNN) previously used ImageNet Large Scale Visual Recognition Challenge (ILSVRC) inversely predict properties single hard magnet (magnetization, size, and location) given 2-D field. show that best network, ResNeXt-50, can perform prediction with error 0.22% magnet.