Novel Machine-Learning Approaches Based on Metric Learning and Manifold Learning for the Design and Knowledge Discovery in Nanophotonics

This talk is focused on using the machine-learning (ML) approaches based on dimensionality reduction and manifold learning for considerably reducing the dimensionality of the forward and inverse problems in relating the input and output of a nanophotonic system. It is shown that by using manifold learning techniques and simplifying the resulting networks using pruning, the computation complexity of the underlying ML algorithms will be considerably reduced. Furthermore, the importance of using an optimally defined loss function (or metric) for ML algorithms in obtaining priceless information about the properties of photonic nanostructures will be emphasized. It is shown that by moving from conventional mean-squared error to more intelligent metrics, knowledge discovery in nanophotonics will be facilitated along with better visualization of the input-output relationship in these nanostructures. In addition to knowledge discovery, the resulting manifold-learning algorithms can be optimally trained to facilitate the inverse design of such nanostructures while minimizing the structural complexity. As such, this talk will provide the foundation for both knowledge discovery and design in photonic nanostructures using manifold learning and metric learning and their application to the highly desired metaphotonic structures as an example platform.