Tuesday , August 20 2019
Home / morocco / This simple piece of glass can recognize the numbers!

This simple piece of glass can recognize the numbers!



[ad_1]

reminiscent of a sci-fi script character for rendering inert of the & # 39 object endowed with intelligence. Nevertheless, it is the researchers of the University of Wisconsin with a glass that can recognize images without electricity source.

You will not look at her glass of beer as he can in a short time was able to spy on you. Researchers from the University of Wisconsin-Madison have managed to create a piece of smart glass, which recognizes the numbers and figures.

Optical neural network to replace the digital neurons

The principle that described Zongfu Yu and his colleagues in the review photonics research based on optical neural networks, where electrons replaced photons (See. Below). The following layers of leaves translucent depart light in a certain direction, which enables to spread the signal in the network to perform the following tasks. startups French Lighton Thus, it developed a "OPU" (optical processing) equivalent processor computer.

Here, researchers have found a way to get rid of these multilayered networks to simplify the process to at least what they call " neural network Nanotechnics ". Small bubbles of different sizes and impurities like graphene have summer included in the glass in certain places. Each bubble or an impurity having index of refraction In particular, it focuses light on a specific location. Light emerging from the image window between one end and the output waves, and then focuses on the exit points 10 on the other side.

Train glass tip as a conventional artificial intelligence

how neural network classic, researchers taught II of, by presenting his collection of 5000 images of numbers from 0 to 9, written by hand in different ways. styleIf light is not focused onto the correct exit point, size, and arrangement of the bubbles have been adjusted to the correct direction signal. After thousands of iterations glass could recognize each figure, including in the forms for which it was not trained. " Each impurity acts as an artificial neuron Zongfu Yu explains that the success of currently limited 79% after 1000 iterations, but it is related to the quality of glass, which is used for the experiment, and it can be improved by material less dense, according to the study.

Biometrics and autonomous vehicles

one of applications It is believed that biometric recognition, For example, to unlock smartphone, Can also be found on the windshield of a Autonomous Vehicles recognize traffic signs. " Since it does not require energyit artificial intelligence It has duration unlimited life, which means that it can provide a device for thousands of years Zongfu Yu says, and it's super-fast, because it runs on … speed of light, " In contrast, a man of vision, a reasonable glass is likely to be reserved for specific applications, Moderate Ming Yuan, professor at the University of Statistics Colombia. A piece of glass in order to recognize the figures, the other – for the determination of the letters, the second – for parties, etc. ".

remember that

  • Researchers from the University of Wisconsin-Madison was able to teach a simple glass to recognize handwritten numbers.
  • This new form of artificial intelligence passive, not requiring energy; it's very fast and economical.
  • Especially it can be used for biometric or autonomous vehicles.

This neural network operates without electricity, the speed of light

article Mark Zaffani published 08/03/2018

Ucla researchers created a neural network of deep learning, which does not work computer with processor classical computation, but with light waves propagates through platelets translucent 3D-prints, in which thousands pixels with relief distracting light.

The fact that only what has been achieved by a research team from the University of California at Los Angeles (UCLA), very exclusive. thanks3D-pressthey created a deep neural network, which does not work with electricity, and with the light. They call it " diffractive deep neural network And to explain the principles of the work in an article published in science.

In the vast world artificial intelligence, Deep Learning (depth training) Currently, it is the most popular technique. It relies on several methodsmachine learning (machine learning) For the training systems of the various types of data with which they will develop a model of representation and abstraction, which would later be used for the interpretation of the information, which they do not know. depth training uses various data layers, thus, the concept of depth.

These layers are complex, audio information is output, which serves as a starting point for the next, and so on. network neurons Artificial multilayer using this architecture. depth training He has made great progress in face recognitionvoice and audio processing language filtering social Networking and analysis of medical images.

Neyranalnaya network of five layers

deep learning systems work on computers all classically. This is where innovation comes Ucla. Their deep neural network needs to electrons, but only in the light. It uses passive components that will perform specific functions, depending on their location. This translucent plate production 3D-press thousands of pixels are coated with a relief. Each plate is a layer of a neural network, the artificial neurons – the pixels that reflect or transmit light at a certain angle.

Researchers used five plyatsovak face to face with the space between them, like dominoes. To test the system, they asked him to recognize the numbers from 0 to 9. The preliminary phase of training of the neural network is obviously carried out on a computer from a computer. database 55000 images of numbers. Researchers have identified the connection created between each layer of artificial neurons and turned them into pixels that are responsible for the management of light in the same way that they then secure with the help of 3D-printing.

Many of the technical constraints remain

In figure was projected during the test laser using a 3D-printing mask placed in front of the first layer. Light passing through the other layers of the neural network, focusing on certain areas. Photodetector at the other end of the network received the output signal light, and then brought back. The system has been tested by thousands of digits, the researchers report that it has reached about 95% accuracy.

While performance is certainly impressive, many technical limitations remain. First, you need a very precise alignment of the platelets to audition light could efficiently reproduce the neural connections. Then, the recognition system is only one of & # 39; an object that must first 3D print in the mask before his project. Modification of the projection device for a full demateryyalizatsyi – this is probably the way to where you can find a system that can find practical application.

Interested in what you just read?

Subscribe to our newsletter daily : Our latest news of the day.

[ad_2]
Source link