Advances in machine learning optimization for classical and quantum photonics

Abstract

The development and optimization of photonic devices and various other nanostructure electromagnetic devices present a computationally intensive task. Much optimization relies on finite-difference time-domain or finite element analysis simulations, which can become very computationally demanding for finely detailed structures and dramatically reduce the available optimization space. In recent years, various inverse design machine learning (ML) techniques have been successfully applied to realize previously unexplored optimization spaces for photonic and quantum photonic devices. In this review, recent results using conventional optimization methods, such as the adjoint method and particle swarm, are examined along with ML optimization using convolutional neural networks, Bayesian optimizations with deep learning, and reinforcement learning in the context of new applications to photonics and quantum photonics.

Non-technological barriers: the last frontier towards AI-powered intelligent optical networks

“…However, compared with conventional physics-based device modeling and design approach, broad adoption of data-driven ML methods is still constrained due to the presence of a number of technical challenges (e.g., the burden of generating gigantic amount of labeled data through simulations and physical experiments to train ML models 5 , risk of unexpected device behavior since ML models even if trained using carefully-assembled data sets may contain singular points at which their predictions can diverge 66 , inflexibility of the ML-based approach since even minor device modifications may require generation of entire training data from the scratch 66 , etc.) as well as various persisting non-technological barriers including explainability and accountability issues 67 , expert workforce limitations, absence of essential standardization frameworks 68 , etc., as mentioned in Table 1 .…”

Exploring Types of Photonic Neural Networks for Imaging and Computing—A Review

“…Particularly compelling is their application in deep learning and pattern recognition, where PNNs harness the parallel processing capabilities of light to execute intricate neural network operations at remarkable speeds, facilitating swift inference and training tasks that strain conventional electronic systems [ 24 , 25 , 26 ]. Moreover, PNNs boast exceptional energy efficiency owing to the minimal losses inherent in photonics, rendering them well-suited for deployment in energy-constrained settings and portable devices.…”