Ensemble nonlinear optical learner by electrically tunable linear scattering

TL;DR

This paper presents an ensemble nonlinear optical learning system using electrically tunable linear scattering in a liquid crystal-polymer film, achieving improved image classification performance with low power and voltage.

Contribution

It introduces a novel ensemble nonlinear optical learner with electrical tunability, enhancing classification accuracy over individual learners in a cost-effective manner.

Findings

Ensemble optical learners outperform individual ones in image classification.

Low optical power and electrical voltage suffice for effective nonlinear optical processing.

Reconfigurability is achieved by simply varying applied voltages.

Abstract

Recent progress in effective nonlinearity, achieved by exploiting multiple scatterings within the linear optical regime, has been demonstrated to be a promising approach to enable nonlinear optical processing without relying on actual material nonlinearity. Here we introduce an ensemble nonlinear optical learner, via electrically tunable linear scattering in a liquid-crystal-polymer composite film under low optical power and low applied electrical voltages. We demonstrate, through several image classification tasks, that by combining inference results from an ensemble of nonlinear optical learners realized at different applied voltages, the ensemble optical learning significantly outperforms the classification performance of individual processors. With very low-level optical power and electrical voltage requirements, and ease in reconfiguration simply by varying applied voltages, the ensemble nonlinear optical learning offers a cost-effective and flexible way to improve computing performance and enhance inference accuracy.