Optical Machine Learning: Harnessing Multiple Scattering for Efficiency

This lecture discusses the innovative use of optics in machine learning, particularly focusing on large-scale random matrix multiplication through multiple scattering of light. The instructor explains how traditional optical machine learning has been limited to small, single-layer networks and simple tasks. By leveraging the complex behavior of light in scattering media, the lecture presents a method to perform essential operations in machine learning optically. The discussion includes the advantages of optical computing, such as low energy consumption and high speed, while also addressing challenges like the need for non-linearities in deep learning networks. The instructor highlights experimental setups and results demonstrating the effectiveness of optical random projections in various machine learning tasks, including image classification and chaotic time series prediction. The lecture concludes with perspectives on future developments in optical computing and its potential convergence with integrated photonics and quantum resources.