Integration of Programmable Diffraction with Digital Neural Networks,ACS Photonics

当前位置： X-MOL 学术 › ACS Photonics › 论文详情

Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)

Integration of Programmable Diffraction with Digital Neural Networks
ACS Photonics ( IF 6.5 ) Pub Date : 2024-08-12 , DOI: 10.1021/acsphotonics.4c01099
Md Sadman Sakib Rahman _{1,

2,

3} , Aydogan Ozcan _{1,

2,

3}

Affiliation

Optical imaging and sensing systems based on diffractive elements have seen massive advances over the last several decades. Earlier generations of diffractive optical processors were, in general, designed to deliver information to an independent system that was separately optimized, primarily driven by human vision or perception. With the recent advances in deep learning and digital neural networks, there have been efforts to establish diffractive processors that are jointly optimized with digital neural networks serving as their back-end. These jointly optimized hybrid (optical + digital) processors establish a new “diffractive language” between input electromagnetic waves that carry analog information and neural networks that process the digitized information at the back-end, providing the best of both worlds. Such hybrid designs can process spatially and temporally coherent, partially coherent, or incoherent input waves, providing universal coverage for any spatially varying set of point spread functions that can be optimized for a given task, executed in collaboration with digital neural networks. In this Perspective, we highlight the utility of this exciting collaboration between engineered and programmed diffraction and digital neural networks for a diverse range of applications. We survey some of the major innovations enabled by the push–pull relationship between analogue wave processing and digital neural networks, also covering the significant benefits that could be reaped through the synergy between these two complementary paradigms.

中文翻译：

可编程衍射与数字神经网络的集成

基于衍射元件的光学成像和传感系统在过去几十年中取得了巨大进步。一般来说，早期几代衍射光学处理器旨在将信息传递到单独优化的独立系统，主要由人类视觉或感知驱动。随着深度学习和数字神经网络的最新进展，人们一直在努力建立与作为后端的数字神经网络联合优化的衍射处理器。这些联合优化的混合（光学+数字）处理器在携带模拟信息的输入电磁波和在后端处理数字化信息的神经网络之间建立了一种新的“衍射语言”，提供了两全其美的效果。这种混合设计可以处理空间和时间上相干、部分相干或不相干的输入波，为任何空间变化的点扩散函数集提供通用覆盖，这些点扩散函数可以针对给定的任务进行优化，并与数字神经网络协作执行。在本视角中，我们强调了工程和编程衍射与数字神经网络之间这种令人兴奋的合作在各种应用中的实用性。我们调查了模拟波处理和数字神经网络之间的推拉关系所带来的一些主要创新，还涵盖了通过这两种互补范式之间的协同作用可以获得的显着好处。

更新日期：2024-08-12

点击分享查看原文

点击收藏

公开下载

阅读更多本刊新发论文本刊介绍/投稿指南