Simple 'smart' glass reveals the future of artificial vision
Smart glass can recognize handwritten digits, even if the writing style is different. The scattered light passes through the glass and then focuses to different locations, which correspond to different digit categories.
The sophisticated technology that powers face recognition in many modern smartphones someday could receive a high-tech upgrade that sounds-and looks-surprisingly low-tech.
Embedding artificial intelligence inside utterly inert objects is a concept that, at first glance, seems like something out of science fiction.
However, it's an advance that could open new frontiers for low-power electronics. Now, artificial intelligence gobbles up substantial computational resources (and battery life) every time you glance at your phone to unlock it with face ID. In the future, one piece of glass could recognize your face without using any power at all.
"This is completely different from the typical route to machine vision." says Yu.
He envisions pieces of glass that look like translucent squares.
But embedded within them, tiny strategically placed bubbles and impurities would bend light in specific ways to differentiate among different images. That's the artificial intelligence in action.
For their proof of concept, the engineers devised a method to make glass pieces that identified handwritten numbers. Light emanating from an image of a number enters at one end of the glass, and then focuses to one of nine specific spots on the other side, each corresponding to individual digits.
The glass was even dynamic enough to detect, in real-time, when a handwritten 3 was altered to become an 8.
"The fact that we were able to get this complex behavior with such a simple structure was really something." says Erfan Khoram, a graduate student in Yu's lab.
Designing the glass to recognize numbers was similar to a machine-learning training process-except that the engineers "trained" an analog material instead of digital codes. Specifically, the engineers placed air bubbles of different sizes and shapes as well as small pieces of light-absorbing materials like graphene at specific locations inside the glass.
"We're accustomed to digital computing, but this has broadened our view." says Yu. "The wave dynamics of light propagation provide a new way to perform analog artificial neural computing."
One such advantage is that the computation is completely passive and intrinsic to the material-and that means one piece of image-recognition glass could be used hundreds of thousands of times.
"We could potentially use the glass as a biometric lock, tuned to recognize only one person's face." says Yu. "Once built, it would last forever without needing power or internet, meaning it could keep something safe for you even after thousands of years."
Additionally, it works at literally the speed of light, because the glass distinguishes among different images by distorting light waves.
Although the up-front training process could be time consuming and computationally demanding, the glass itself is easy and inexpensive to fabricate.
In the future, the researchers plan to determine if their approach works for more complex tasks, such as facial recognition.
"The true power of this technology lies its ability to handle much more complex classification tasks instantly without any energy consumption," says Ming Yuan, a professor of statistics at Columbia University. "These tasks are the key to create artificial intelligence: to teach driverless cars to recognize a traffic signal, to enable voice control in consumer devices, among numerous other examples."
Unlike human vision, which is mind-bogglingly general in its capabilities to discern untold thousands of different objects, the smart glass could excel in specific applications-for example, one piece for number recognition, a different piece for identifying letters, another for faces, and so on.
"We're always thinking about how we provide vision for machines in the future, and imagining application specific, mission-driven technologies." says Yu. "This changes almost everything about how we design machine vision."
Author brief introduction
Zongfu Yu is the Dugald C. Jackson Associate Professor and Vilas Associate at UW-Madison. Graduate students Erfan Khoram, Ang Chen and Dianjing Liu contributed to the research. Qiqi Wang at Massachussetts Institute of Technology and Ming Yuan at Columbia University were collaborators. A DARPA Young Faculty Award program supported the research.
Sam Million-Weaver, perspective@engr.wisc.edu, (608) 263-5988
Yu Research Group: https://photonics.engr.wisc.edu/
简易“智能”玻璃揭示人工视觉的未来
智能玻璃可以识别出手写的数字。即使书写风格迥异，它也可以正确识别。图中显示物体的光经过玻璃被散射然后聚焦到不同的位置，这些位置对应不同的物理类别。
在未来的某天，许多现代智能手机中驱动人脸识别的顶尖技术可能会有一次全面升级。但令人意外的是，这次升级不管是听起来还是看上去都更像是一种“低端技术”。
此次通往未来的窗口仅仅是一块玻璃--来自威斯康星大学麦迪逊分校的工程师们设计了一块“智能”玻璃，这种玻璃可以在不借助任何探测器、电路和电源的情况下识别图像。
“我们正在利用光学方法把照相机、探测器和深度神经网络等集成在单块薄玻璃上。”来自麦迪逊分校电子与计算机工程系的助理教授Zongfu Yu如此说道。相关研究成果作为封面文章发表在Photonics Research第7卷第8期上。(Erfan Khoram, et al., Nanophotonic media for artificial neural inference).
将人工智能嵌入完全非智能的物体中，乍一看更像是来自科幻小说中的故事。然而，这却是一种可以开拓低功耗电子学新领域的方式。现在，每当你利用面部ID解锁手机时，人工智能就会消耗大量计算资源(以及电池寿命)。而在未来，一小块玻璃就可以达到相同的目的，并且不消耗任何电源。
“这完全不同于机器视觉的典型路线。”Yu说。他设想该玻璃看起来就像是半透明的方块，里面嵌着精心放置的气泡和杂质，可以以特定方式弯曲光线来区分不同图像。这正是人工智能起的作用。
具体到概念验证方面，工程师们设计了一种制作玻璃块的方法，可以识别手写数字。来自数字图像的光从玻璃的一端发出，然后会聚在另一端十个斑点中的一个，这十个斑点分别对应十个数字。该玻璃甚至可以实现实时的动态探测，例如从手写数字3演变成8。
“利用一种简单的结构来实现如此复杂的效果，这个结果令人振奋。”Yu实验室的研究生Erfan Khoram表示。
设计玻璃来识别数字的过程与机器学习的训练过程非常类似，所不同的是，工程师们“训练”的是一块模拟材料而非数字代码。具体来说，他们在一层仅为20 μm厚的薄玻璃中，将大小不同、形状迥异的空气气泡和小片吸光材料(如石墨烯)放置在设计好的位置上。
不同于具有层状结构的数字神经网络，该玻璃以一种连续的方式来进行神经计算。所有神经计算中最为关键的要素就是非线性激励函数，这里是通过嵌入小片的光学可饱和吸收体(如石墨烯)来实现的。
“人们通常习惯于数字计算，但这种方法拓展了我们的视野。”Yu说，“光传播的波动力学提供了一种进行类似人工神经计算的新方法。”
这种方法的一个优点是，计算是完全无源的，并且是材料的固有属性，这就意味着一块图像识别玻璃可以重复使用成千上万次。
“该玻璃可以当作一种生物识别锁，在调谐之后可以只识别一个人的脸。”Yu说，“这种联系一旦建立，在没有电源和网络的情况下也能发挥作用。这意味着即使在千万年之后也能对你的某个事物提供安全保障。”
此外，由于该玻璃是通过扭曲光线来区别不同图片的，所以从理论上说它可以以光速工作。
虽然前期的训练过程需要大量的时间和计算量，但是玻璃的制作工艺简单，成本低廉。
未来，学者们计划验证该方法能否适用于更复杂的任务，如人脸识别。
“这项技术的真正威力在于它能够在不消耗任何电源的情况下即时处理复杂的分类任务。”哥伦比亚大学统计学教授Ming Yuan表示，“这些任务是创造人工智能的关键，例如让无人驾驶汽车识别交通信号，在消费类设备中实现语音控制等等。”
在识别不同物体时，人类视觉表现出了几乎相同的识别能力。而这种智能玻璃更擅长识别特定的事物，如数字、字母、人脸等等。
“我们一直在思考如何给未来的机器提供视觉和特定成像应用方面的视觉感知。”Yu说，“这种设计几乎颠覆了传统的机器视觉设计方式。”
作者简介
Zongfu Yu是麦迪逊分校的Dugald C. Jackson副教授和Vilas Associate。研究生Erfan Khoram, Ang Chen和Dianjing Liu对该研究有贡献。麻省理工学院的Qiqi Wang和哥伦比亚大学的Ming Yuan是该研究的合作者。DARPA Young Faculty Award program 支持了该项研究。
Yu 研究小组主页：https://photonics.engr.wisc.edu/