Interfacing photonics with artificial intelligence: a new design strategy for photonic metamaterials based on deep learning

Over the past decades, we have witnessed tremendous progress and success of photonic metamaterials. By tailoring the geometry of the building blocks of metamaterials and engineering their spatial distribution, we can control the amplitude, polarization state, phase and trajectory of light in an almost arbitrary manner. However, the conventional physics- or rule-based approaches are insufficient for designing multi-functional and multi-dimensional metamaterials, since the degrees of freedom in the design space become extremely large. Deep learning, a subset of machine learning that learns multilevel abstraction of data using hierarchically structured layers, could potentially accelerate the development of complex metamaterials and other photonic structures with high efficiency, accuracy and fidelity. In this talk, I will present our recent works that employ advanced deep learning techniques to design and evaluate distinct photonic metamaterials.