Collaborative Research: CNS Core: Medium: TeTON: A Testbed and a Toolkit for Expediting Investigation of and Accelerating Advancements in All-Optical Neural Networks

在大脑中，数十亿个神经元不断收集彼此的电信号，通过这些信号的自然非线性组合，人类能够处理信息(视觉、音频、触觉等)，学习如何识别模式，并做出快速决定。这些相互连接的神经元形成了通常所说的神经网络。该项目旨在开发全光(人工)神经网络(AONN)试验台和相关软件工具包，以加速和加速光学人工智能的进步，包括具有自动机器学习能力的可编程多层AONN硬件和用于公共教育的开源软件包。AONN试验台建立在自由空间傅里叶光学和超弱光级非线性光学的基础上，可以有效地重建光子功能，类似于在人脑中执行的功能。利用光的电磁波特性，与目前已被应用于实现人工智能(AI)的更传统的软件和定制的基于电子的人工神经网络相比，所产生的AONN有望运行得更快，消耗的能源明显更少。发现如何最好地、最有效地实施AONN解决方案将改变实现AI的方式。能够以光速思考的人工大脑可以为实现许多新的、目前无法想象的实时应用铺平道路，例如，检查大量制造对象的质量，并在需要快速调整生产时向生产工厂提供几乎即时的反馈。使用AONN能够快速识别模式及其异常，将在医学、教育、制造、交通、农业、自然科学等领域找到许多其他应用。下面的例子可以说明实现和使用AONN的潜在优势。AONN可以在不到一毫秒的时间内对运动员或患者在执行特定任务时进行康复的摄像头生成的每个视频帧进行单独分析。AONN可以很容易地检测到不正确或不理想的身体动作，从而向运动员或患者提供即时反馈，他们反过来可以不断纠正他们的姿势，以实现更好的结果。该项目支持本科生和研究生的职业发展，他们共同致力于这个跨学科的项目。得克萨斯大学达拉斯分校(UTD)和堪萨斯大学(KU)开发了课程和推广材料，并与公众分享，重点是与来自代表不足的社区的学生接触，并为未来STEM学科的学生多样性做出贡献。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

In a brain, billions of neurons continuously collect electrical signals from one another and through natural non-linear combination of these signals human beings are able to process information (visual, audio, tactile, etc.), learn how to recognize patterns, and make quick decisions. These interconnected neurons form what is commonly known as a neural network. This project aims to develop an all-optical (artificial) neural network (AONN) testbed and related software toolkit for expediting and accelerating advancements in optical artificial intelligence, including a programmable multi-layer AONN hardware with automatic machine learning capability and an open-source software package for public education. The AONN testbed builds upon free-space Fourier optics and ultra-low light level nonlinear optics, which can efficiently recreate with photons functionalities similar to those that are performed in the human brain. Making use of the electromagnetic wave nature of light the resulting AONN is expected to run much faster and consume significantly less energy when compared to the more conventional software and custom electronic-based artificial neural networks that have been applied to realize artificial intelligence (AI) so far.Discovering how to best and most effectively implement AONN solutions is going to change the way in which AI is achieved. An artificial brain that can think at the speed-of-light can pave the way to the realization of a number of new and currently unthinkable real-time applications, like for example inspecting the quality of a large number of manufactured objects and providing virtually instantaneous feedback to the production plant in the event that swift production adjustments are required. Being able to quickly recognize patterns and their anomalies using AONN will find many other applications in the fields of medicine, education, manufacturing, transportation, agriculture, natural science, etc. The following example may illustrate the potential advantages of implementing and using AONN. Each video frame generated by a camera observing an athlete or a patient going though rehabilitation while performing specific tasks could be individually analyzed by AONN in less than one millisecond. Improper or suboptimal body motions would be readily detected by AONN, hence providing instantaneous feedback to the athlete or patient who in turn could continuously correct their posture to achieve improved outcomes. This project supports career development of the undergraduate and graduate students, who work together on this cross-disciplinary project. Courses and outreach materials are developed at the University of Texas at Dallas (UTD) and the University of Kansas (KU) and shared with the public, with an emphasis on engaging with students from underrepresented communities and contributing to future student diversity in STEM disciplines.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.