Data-efficient Modeling of Optical Matrix Multipliers Using Transfer Learning

TL;DR

This paper presents a transfer learning approach to efficiently model optical matrix multipliers, significantly reducing data requirements and outperforming traditional analytical models in experimental settings.

Contribution

It introduces a transfer learning-based method for modeling optical matrix multipliers that requires less than 10% of the data typically needed, improving accuracy over analytical models.

Findings

Requires less than 10% of data for effective modeling

Outperforms analytical models in experimental tests

Demonstrates data efficiency in optical neural network components

Abstract

We demonstrate transfer learning-assisted neural network models for optical matrix multipliers with scarce measurement data. Our approach uses <10\% of experimental data needed for best performance and outperforms analytical models for a Mach-Zehnder interferometer mesh.