Class-specific diffractive cameras based on deep learning-designed surfaces

Abstract

Recently, a new diffractive camera design based on transmissive surfaces structured using deep learning is proposed. It performs class-specific imaging of target objects and all-optical deletion of other classes of objects, which will promote the development of privacy-preserving digital cameras and mission-specific data.

Backpropagation-free training of deep physical neural networks

“…. Such trainable matrix multiplications can be performed either digitally or through physical systems, for instance using Mach-Zehnder interferometer (MZI)-integrated photonics (43) or spatial light modulators (SLMs) in optics (21,44). The goal is to train W l ð Þ t locally without the need to know the nonlinear physical layer.…”

“…For details on data generation, refer to supplementary text, section 2.2. digit, fashion Mnist, and CIFAR10, based on three distinct physical systems, each featuring a distinctive source of nonlinearity (materials and methods and supplementary text, sections 2.3 to 2.5). Although there have been proposals that explore wave-based analog computing for linear operations, such as multiplication and convolution (43,(45)(46)(47)(48)(49)(50)(51)(52)(53), it is important to note that PNNs require nonlinearity to effectively handle regression and classification tasks. We evaluated the performance of PhyLL in these media (refer to tables S1 and S2 in the supplementary materials) against in silico and BP methods under both supervised and unsupervised contrastive training schemes using an end-to-end surrogate forward model of the systems for benchmarking purposes.…”

Hollow-core optical fibers: current state and development prospects

“…The pump radiation from a thulium fiber laser with 8 W of power, tunable within a spectral range near 1.96 mm, was introduced into a 5-m-long hollow-core filled with HBr molecules at a pressure of 5 mbar. As a result, CW laser oscillation with stepwise wavelength tuning within the range of 3.81±4.496 mm was implemented [168]. The GFL output power reached 500 mW, the efficiency being about 18%.…”

Untitled

“…The focus on size, weight, power, price, and performance (SWaP3) optimization in LWIR thermal imaging systems is a key trend in research and development. Metalenses, capable of their subwavelength scale manipulation of wavefront information such as amplitude, phase, and polarization, mark a significant advancement in optical technology [4][5][6][7][8][9][10] . They offer the potential to replace or supplement traditional lenses in applications requiring high-precision imaging or in miniature optical systems, attracting significant attention recently.…”