Photonic extreme learning machine by free-space optical propagation

Abstract

Photonic brain-inspired platforms are emerging as novel analog computing devices, enabling fast and energyefficient operations for machine learning. These artificial neural networks generally require tailored optical elements, such as integrated photonic circuits, engineered diffractive layers, nanophotonic materials, or time-delay schemes, which are challenging to train or stabilize. Here, we present a neuromorphic photonic scheme, i.e., the photonic extreme learning machine, which can be implemented simply by using an optical encoder and coherent wave propagation in free space. We realize the concept through spatial light modulation of a laser beam, with the far field acting as a feature mapping space. We experimentally demonstrate learning from data on various classification and regression tasks, achieving accuracies comparable with digital kernel machines and deep photonic networks. Our findings point out an optical machine learning device that is easy to train, energetically efficient, scalable, and fabrication-constraint free. The scheme can be generalized to a plethora of photonic systems, opening the route to real-time neuromorphic processing of optical data.

Parallel Extreme Learning Machines Based on Frequency Multiplexing

“…When executed with 784 input neurons, 784 hidden neurons, and a tanh nonlinearity, this ELM reached an accuracy of 92.8% (the accuracy could be increased up to approximately 98%, increasing the number of hidden neurons by 20 times). We report experimental results compared with the numerical simulation of our system, the score obtained by another photonic ELM reported in [8], and a software ELM with a number of neurons comparable to our experiment. The right panel contains results from the MNIST classification task.…”

“…The left panel contains results from the mushroom classification task. We report experimental results compared with the numerical simulation of our system, the score obtained by another photonic ELM reported in[8], and a software ELM with a number of neurons comparable to our experiment. The right panel contains results from the MNIST classification task.…”

“…A photonic ELM exploiting free-space propagation in a scattering medium [8] scored an accuracy of 95.4% on this same task.…”

“…Different schemes for photonic implementations of randomized neural networks have already been explored. ELMs have been implemented in setups based on free-space propagation along scattering media [7,8], multimode fibers [9], or time-multiplexed fiber loops [10]. Many of these substrates have also been exploited for Reservoir Computing (RC), which is a randomized approach to recursive neural networks.…”

Artificial Intelligence in Classical and Quantum Photonics

“…In the last years, research on photonic computation, [ 217 ] neuromorphic engineering [ 218–220 ] and RC are growing. [ 207,221,222 ] Along this research path, new application of photonic hardware and especially SLMs have been demonstrated, such as computing the ground state of systems of interacting spins [ 223,224 ] or perform classical ML task leveraging similar computational frameworks such as extreme learning machines [ 225 ] exploiting light propagation in free‐space [ 226 ] or through fibers. [ 227 ]…”