In
recent years, artificial Intelligence has remarkably accelerated the
developments of many realms of science and engineering, such as material
science, chemistry, physics, quantum mechanics, and computational
imaging. Artificial intelligence has also made tremendous advances in
the fields of nanophotonics, metamaterials and metasurfaces, allowing
efficient forward modeling and inverse design. This tutorial will focus
on applications of artificial intelligence in nanophotonics,
metamaterials, and metasurfaces. The tutorial will be divided into four
parts:
An Overview of On-Demand Design of Metamaterials Enabled by Deep Learning
Yongmin Liu, Northeastern University
Over
the last years, deep learning has been used for on-demand design of
metamaterials by representing and learning the mapping between the
topology and composition of metamaterials and their associated optical
properties. Liu will review the recent progress in on-demand design of
metamaterials using deep learning, with an emphasis on various model
architectures for specific photonic tasks. He will also provide the
historical background, key algorithms, fundamentals, and applications of
deep learning used for design of metamaterials.
Decoding Optical Data with Machine Learning
Yuebing Zheng, The University of Texas at Austin
With
the rapid development of various optical spectroscopy and imaging
techniques, accurate and rapid analysis of optical data is highly
demanding. Meanwhile, machine learning has shown its exceptional
capability of decoding complex data over the last decade. Zheng will
provide the recent progress in machine learning assisted decoding
optical data, which allows accurate analysis of optical data. He will
focus on their applications in a wide range of fields.
An Overview of Numerical Optimization Methods for Metasurfaces
Patrice Genevet, Centre de Recherche sur I’Hétéro- Epitaxie et ses Applications
Numerical
optimizations search for local or global optimal solutions by either
evolutionary or gradient-based approaches. The first method searches
solutions stochastically, in an extensive parameter space allowing
capture of global optima; the second strategy calculates the derivatives
of the cost function and finds local optima. Genevet will review the
recent advanced optimization method to further exploit metasurface
capabilities. He will also discuss Bayesian-based optimization
techniques which can reduce the computational cost substantially.
An Overview of Machine Learning-Assisted Global Optimization of Photonic Devices
Alexandra Boltasseva, Purdue University
Machine
learning has shown great promise for generating data by learning
important features. By exploiting its versatility and efficiency,
machine-learning assisted novel optimization methods have been reported,
coupled with conventional optimization framework including topology
optimization and meta-heuristic optimization. In her tutorial,
Boltasseva will present the recent progress in global optimization
assisted by machine learning, with an emphasis on high efficiency and
substantial improvement.