Photonic neural networks with spatiotemporal chaos in multimode fibers

Bahadır Utku Kesgin,Uğur Teğin

2024-11-01

Abstract:Optical computing has gained significant attention as a potential solution to the growing computational demands of machine learning, particularly for tasks requiring large-scale data processing and high energy efficiency. Optical systems offer promising alternatives to digital neural networks by exploiting light's parallelism. This study explores a photonic neural network design using spatiotemporal chaos within grad-ed-index multimode fibers to improve machine learning performance. Through numerical simulations and experiments, we show that chaotic light propagation in multimode fibers enhances data classification accu-racy across domains, including biomedical imaging, fashion, and satellite geospatial analysis. This chaotic optical approach enables high-dimensional transformations, amplifying data separability and differentiation for greater accuracy. Fine-tuning parameters such as pulse peak power optimizes the reservoir's chaotic properties, highlighting the need for careful calibration. These findings underscore the potential of chaos-based nonlinear photonic neural networks to advance optical computing in machine learning, paving the way for efficient, scalable architectures.

Optics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to utilize the spatiotemporal chaos phenomenon in photonic neural networks to improve the accuracy of data classification in machine - learning tasks. Specifically, the researchers explored the introduction of spatiotemporal chaos dynamics in graded - index multimode fibers (GRIN MMFs) to enhance the performance of fiber - based photonic neural network architectures. Through numerical simulations and experimental studies, they demonstrated that chaotic light propagation can significantly improve the classification accuracy on multiple datasets, including datasets such as Breast MNIST (breast ultrasound images), Fashion MNIST (fashion clothing pictures), and EuroSAT (satellite geospatial analysis). This method increases the dimension and separability of the model through high - order mode excitation and nonlinear effects, thereby improving the performance of machine - learning tasks.

The key to the research lies in optimizing parameters such as pulse peak power to maintain high - efficiency machine - learning task processing capabilities. In addition, this study also explored the advantages of this spatiotemporal - chaos - based photonic neural network architecture in terms of energy consumption and its ability to enhance the linear separability of data in the feature space. These findings open up new avenues for the development of sustainable and scalable optical computing hardware.