2021-072 Ensemble Learning of Diffractive Optical Networks

SUMMARY

UCLA researchers in the Department of Electrical and Computer Engineering have improved the statistical inference performance of diffractive optical networks for artificial intelligence-related applications.

BACKGROUND

Optical computing platforms are having renewed interest in artificial intelligence-related applications. Due to their high speed, large bandwidth and high interconnectivity of optical information processing, they are well suited for realizing neural network models. Despite these advantages, optical computing frameworks have had limited application due to their poor inference performance. By improving the inference performance, diffractive optical systems can expand the abilities of deep neural networks (DNNs) to realize miniaturized, ultrafast machine learning solutions for a variety of applications, including all-optical object classification, diffraction-based optical computing hardware, and computational imaging tasks

INNOVATION

UCLA researchers in the Department of Electrical and Computer Engineering developed Diffractive Deep Neural Networks (D²NNs), an optical computing framework that improves inference performance of diffractive optical networks by using feature engineering and ensemble learning. D²NN is comprised of successive transmissive and/or reflective diffractive surfaces that can process input information through light-matter interaction. The surfaces are designed using standard deep learning techniques and fabricated and assembled into a physical optical network. Together with the advances in the fabrication and assembly of nanoscale optical systems, D²NN closes the performance gap between optical neural networks and their digital counterparts.

POTENTIAL APPLICATIONS

• Miniaturized, ultrafast machine learning solutions
• All-optical object classification
• Diffraction-based optical computing hardware
• Computational imaging tasks
• Diffractive optical neural network designs
• Optical microscopy
• Parallelism of optical systems
• Optics-based computation
• Machine learning
• Object classification, spectral-encoding of information, optical pulse shaping and imaging

ADVANTAGES

• Highest inference accuracies achieved to date by any diffractive optical neural network design on the same dataset
• Uses feature engineering and ensemble learning to improve the inference performance of diffractive optical networks
• Improvement on measurements
• Parallel processing of optical information
• Image classification accuracy
• Individual diffractive classifiers

RELATED MATERIALS

• Rahman, M.S.S., Li, J., Mengu, D. et al. Ensemble learning of diffractive optical networks. Light Sci Appl 10, 14 (2021). https://doi.org/10.1038/s41377-020-00446-w

STATUS OF DEVELOPMENT

The researchers evaluated the performance of the D²NN ensembles on CIFAR-10 image dataset containing 60,000 natural images categorized in 10 classes. Simulations of an ensemble of 14 individually trained D²NNs achieved 61.21% blind testing accuracy on CIFAR-10 dataset, ~16% higher than the average accuracy of the individual constituent D²NNs.