Fig. 2: Neural network performance analysis.
From: Large area optimization of meta-lens via data-free machine learning
a An example field. (left) is the absolute value of the ground truth finite difference frequency ___domain (FDFD) simulated field, (middle) difference between the true value of the field and the field produced by using the local phase approximation (LPA), and (right) is the absolute difference between the neural network field and the FDFD field. b Comparison between the performance of the proposed neural network and LPA methods. (left) Shows the relative error between FDFD predicted fields and the fields predicted by LPA. (right) Error comparisons between the transmission coefficients predicted by LPA and the neural net.
