Introducing a dynamic deep neural network to infer lens design starting points

摘要

Most lens design problems involve the time-consuming task of finding a proper starting point, that is, a lensdesign that approximately fulfills the desired first-order specifications while decently correcting aberrations. Inrecent work, a fully-connected (FC) deep neural network was trained to learn this task by extrapolating fromknown lens design databases. Here, we introduce a new dynamic neural-network architecture for the startingpoint problem which is based on a recurrent neural network (RNN) architecture. As we show, the dynamicnetwork can learn to infer good starting points on many lens design structures at once whereas the previousmodel was limited to a given sequence of glass elements and air gaps. We also show that a pretrained RNNmodel can generalize its knowledge over new lens design structures for which we have no reference lens designand obtain a significantly better optical performance than a RNN trained from scratch.