Dataset Information

Training large-scale optoelectronic neural networks with dual-neuron optical-artificial learning.

ABSTRACT: Optoelectronic neural networks (ONN) are a promising avenue in AI computing due to their potential for parallelization, power efficiency, and speed. Diffractive neural networks, which process information by propagating encoded light through trained optical elements, have garnered interest. However, training large-scale diffractive networks faces challenges due to the computational and memory costs of optical diffraction modeling. Here, we present DANTE, a dual-neuron optical-artificial learning architecture. Optical neurons model the optical diffraction, while artificial neurons approximate the intensive optical-diffraction computations with lightweight functions. DANTE also improves convergence by employing iterative global artificial-learning steps and local optical-learning steps. In simulation experiments, DANTE successfully trains large-scale ONNs with 150 million neurons on ImageNet, previously unattainable, and accelerates training speeds significantly on the CIFAR-10 benchmark compared to single-neuron learning. In physical experiments, we develop a two-layer ONN system based on DANTE, which can effectively extract features to improve the classification of natural images.

SUBMITTER: Yuan X

PROVIDER: S-EPMC10625607 | biostudies-literature | 2023 Nov

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Training large-scale optoelectronic neural networks with dual-neuron optical-artificial learning.

Yuan Xiaoyun X Wang Yong Y Xu Zhihao Z Zhou Tiankuang T Fang Lu L

Nature communications 20231104 1

Optoelectronic neural networks (ONN) are a promising avenue in AI computing due to their potential for parallelization, power efficiency, and speed. Diffractive neural networks, which process information by propagating encoded light through trained optical elements, have garnered interest. However, training large-scale diffractive networks faces challenges due to the computational and memory costs of optical diffraction modeling. Here, we present DANTE, a dual-neuron optical-artificial learning ...[more]

PMID: 37925451

Similar Datasets

Project description:Objective.Non-human primates (NHPs) are critical for development of translational neural technologies because of their neurological and neuroanatomical similarities to humans. Large-scale neural interfaces in NHPs with multiple modalities for stimulation and data collection poise us to unveil network-scale dynamics of both healthy and unhealthy neural systems. We aim to develop a large-scale multi-modal interface for NHPs for the purpose of studying large-scale neural phenomena including neural disease, damage, and recovery.Approach.We present a multi-modal artificial dura (MMAD) composed of flexible conductive traces printed into transparent medical grade polymer. Our MMAD provides simultaneous neurophysiological recordings and optical access to large areas of the cortex (∼3 cm2) and is designed to mitigate photo-induced electrical artifacts. The MMAD is the centerpiece of the interfaces we have designed to support electrocorticographic recording and stimulation, cortical imaging, and optogenetic experiments, all at the large-scales afforded by the brains of NHPs. We performed electrical and optical experiments bench-side andin vivowith macaques to validate the utility of our MMAD.Main results.Using our MMAD we present large-scale electrocorticography from sensorimotor cortex of three macaques. Furthermore, we validated surface electrical stimulation in one of our animals. Our bench-side testing showed up to 90% reduction of photo-induced artifacts with our MMAD. The transparency of our MMAD was confirmed both via bench-side testing (87% transmittance) and viain vivoimaging of blood flow from the underlying microvasculature using optical coherence tomography angiography.Significance.Our results indicate that our MMAD supports large-scale electrocorticography, large-scale cortical imaging, and, by extension, large-scale optical stimulation. The MMAD prepares the way for both acute and long-term chronic experiments with complimentary data collection and stimulation modalities. When paired with the complex behaviors and cognitive abilities of NHPs, these assets prepare us to study large-scale neural phenomena including neural disease, damage, and recovery.

Dataset Information

Training large-scale optoelectronic neural networks with dual-neuron optical-artificial learning.

Publications

Training large-scale optoelectronic neural networks with dual-neuron optical-artificial learning.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets