Training neural networks with end-to-end optical backpropagation

Abstract

Optics is an exciting route for the next generation of computing hardware for machine learning, promising several orders of magnitude enhancement in both computational speed and energy efficiency. However, to reach the full capacity of an optical neural network it is necessary that the computing not only for the inference, but also for the training be implemented optically. The primary algorithm for training a neural network is backpropagation, in which the calculation is performed in the order opposite to the information flow for inference. While straightforward in a digital computer, optical implementation of backpropagation has so far remained elusive, particularly because of the conflicting requirements for the optical element that implements the nonlinear activation function. In this work, we address this challenge for the first time with a surprisingly simple and generic scheme. Saturable absorbers are employed for the role of the activation units, and the required properties are achieved through a pump-probe process, in which the forward propagating signal acts as the pump and backward as the probe. Our approach is adaptable to various analog platforms, materials, and network structures, and it demonstrates the possibility of constructing neural networks entirely reliant on analog optical processes for both training and inference tasks.

References in corpus (9)

• Photonics for artificial intelligence and neuromorphic computing
• 11 TeraFLOPs per second photonic convolutional accelerator for deep learning optical neural networks
• Deep physical neural networks enabled by a backpropagation algorithm for arbitrary physical systems
• Large-scale neuromorphic optoelectronic computing with a reconfigurable diffractive processing unit
• Experimentally realized in situ backpropagation for deep learning in nanophotonic neural networks
• Delocalized Photonic Deep Learning on the Internet's Edge
• Single chip photonic deep neural network with accelerated training
• Silicon Photonic Architecture for Training Deep Neural Networks with Direct Feedback Alignment
• Fully reconfigurable coherent optical vector-matrix multiplication

Cited by in corpus (4)

• Asymmetrical estimator for training encapsulated deep photonic neural networks
• Engineering nonlinear activation functions for all-optical neural networks via quantum interference
• Quantum optical shallow networks
• Streamlined optical training of large-scale modern deep learning architectures with direct feedback alignment