High-Dimensional Optical Nearest Centroid Classification Using Orbital Angular Momentum Modes of Light

Abstract

Distance evaluation between high-dimensional vectors is a fundamental computational primitive in classical machine learning, widely employed to quantify similarity, yet often demanding substantial computational resources. The quantum nearest-centroid classifier leverages the principle of superposition to encode data vectors, enabling direct distance evaluation through inner-product measurements in a high-dimensional state space. In this work, we propose and experimentally demonstrate a quantum-inspired optical nearest-centroid (ONC) classifier that leverages the spatial superposition of high-dimensional orbital angular momentum (OAM) modes in classical light to optically encode data vectors and perform distance evaluation. We first experimentally characterize the measurement accuracy of the core distance estimation module across 32 OAM modes. Subsequently, we demonstrate its performance on real-world data, achieving high-dimensional classification using the four-dimensional Iris data set and a 16-dimensional subset of the complex MNIST data set. Experimental results show excellent agreement with theoretical expectations. Our work demonstrates that classical OAM modes can efficiently emulate complex quantum-inspired algorithms, establishing a scalable optical platform for resource-efficient high-dimensional data processing.