Incoherent Light-Driven Nonlinear Optical Extreme Learner via Data Reverberation

TL;DR

This paper introduces a low-power, incoherent-light-driven optical extreme learner that uses data reverberation in a tailored optical cavity to achieve nonlinear transformations, outperforming linear digital networks in image classification and XOR tasks.

Contribution

The work presents a novel optical neural network approach that eliminates the need for nonlinear materials by leveraging data reverberation, enabling scalable, energy-efficient optical machine learning.

Findings

Achieves nonlinear transformations at extremely low optical power.

Outperforms linear digital networks in classification tasks.

Matches accuracy of fully nonlinear digital models.

Abstract

Artificial neural networks have revolutionized fields from computer vision to natural language processing, yet their growing energy and computational demands threaten future progress. Optical neural networks promise greater speed, bandwidth, and energy efficiency, but suffer from weak optical nonlinearities. Here, we demonstrate a low-power, incoherent-light-driven optical extreme learner that leverages 'data nonlinearity' from optical pattern reverberation, eliminating reliance on intrinsic nonlinear materials. By encoding input data in the spatial polarization distribution of a tailored optical cavity and allowing light to pass through it multiple times, we achieve nonlinear transformations at extremely low optical power. Coupled with a simple trainable readout, our optical learner consistently outperforms linear digital networks in standard image classification tasks and XOR benchmarks, delivering accuracy matching fully nonlinear digital models. Our compact, energy-efficient approach significantly reduces complexity, cost, and energy consumption, paving the way for practical, scalable all-optical machine learning platforms.