English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Analog physical neural networks, which hold promise for improved energyefficiency and speed compared to digital electronic neural networks, arenevertheless typically operated in a relatively high-power regime so that thesignal-to-noise ratio (SNR) is large (>10). What happens if an analog system isinstead operated in an ultra-low-power regime, in which the behavior of thesystem becomes highly stochastic and the noise is no longer a smallperturbation on the signal? In this paper, we study this question in thesetting of optical neural networks operated in the limit where some layers useonly a single photon to cause a neuron activation. Neuron activations in thislimit are dominated by quantum noise from the fundamentally probabilisticnature of single-photon detection of weak optical signals. We show that it ispossible to train stochastic optical neural networks to perform deterministicimage-classification tasks with high accuracy in spite of the extremely highnoise (SNR ~ 1) by using a training procedure that directly models thestochastic behavior of photodetection. We experimentally demonstrated MNISTclassification with a test accuracy of 98% using an optical neural network witha hidden layer operating in the single-photon regime; the optical energy usedto perform the classification corresponds to 0.008 photons permultiply-accumulate (MAC) operation, which is equivalent to 0.003 attojoules ofoptical energy per MAC. Our experiment used >40x fewer photons per inferencethan previous state-of-the-art low-optical-energy demonstrations, to achievethe same accuracy of >90%. Our work shows that some extremely stochastic analogsystems, including those operating in the limit where quantum noise dominates,can nevertheless be used as layers in neural networks that deterministicallyperform classification tasks with high accuracy if they are appropriatelytrained.

Quantum-noise-limited optical neural networks operating at a few quanta per activation