Compressing and expanding optical matrix-vector multipliers for enabling optical image encoder-decoders and generators

TL;DR

This paper introduces a new optical multiplier design to enable full optical neural networks for image processing tasks.

Contribution

The novel expanding optical matrix-vector multiplier complements compressing multipliers for optical neural networks.

Findings

Expanding and compressing optical multipliers are combined to build image processor networks.

The design enables optical autoencoders and image generators.

Full optical realization of neural networks is made possible with the proposed multiplier schemes.

Abstract

Both compressing and expanding optical matrix-vector multipliers are necessary for the full optical realization of neural networks. An expanding multiplier scheme is proposed, which, together with common compressing multipliers, is employed to demonstrate image processor networks such as autoencoders and image generators.

Figures