Our everyday beings are potential computers

Physicists made artificial neural networks by combining objects instead of computer chips. They work by taking advantage of physical properties inherent in mechanical systems, such as the vibration of a metal plate.

A frying pan, a sandpaper, a beaker… These things that surround us can run artificial intelligence programs, much more efficient than the traditional chips in our computers.

Anyway, that’s what they imagine Logan Wright, a physicist at Cornell Universityand colleagues in the United States. They showed it in Study published on January 26 in a temper nature. in a Science, Researcher says:

Anything can turn into a computer. We are just beginning to find ways to exploit the physical properties of materials to our liking.”

For the study’s authors, the body’s response to a stimulus is a passive method of “computing”, unlike semiconductors, which use electricity to perform millions or even billions of calculations. What must move a chess piece correctly on a computer screen. after every thing, “The grooves add a lot of our echoes without having to use specific computer software,” pointed out Science.

With this idea in mind, and to prove their concept, the researchers created several neural networks – a type of artificial intelligence that loosely simulates the brain to perform complex tasks – by combining objects. They then exploited their intrinsic properties in three systems: an optical system, by passing light through a crystal, a mechanical system, by vibrating a metal plate, and an electronic system, by injecting current through small circuits.

Trained to recognize numbers accurately

Next, they “trained” these systems, called “physical neural networks”, to perform tasks, for example recognizing handwritten numbers for one or recognizing the sounds of another. “The accuracy of these tasks ranges from 87 to 97%,” They reported in the study.

See also  Is your computer lagging? Do a big clean up to save space!

For the authors, this work shows that these physical neural networks have the potential to perform machine learning faster and more energy efficient than traditional electronic processors.

source

Prestigious magazine created in 1848. It provides a particularly detailed and comprehensive panoramic view of the country and science debates in the United States and in the rest of the world. The site deals with the topics presented in the weekly newspaper and places them

[…]

read more

Frank Mccarthy

<p class="sign">"Certified gamer. Problem solver. Internet enthusiast. Twitter scholar. Infuriatingly humble alcohol geek. Tv guru."</p>

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top