A technology for artificially creating a rare form of matter known as Spin Glass promises to create a new paradigm in artificial intelligence, allowing algorithms to directly imprint physical devices.
The extraordinary properties of these materials make it possible to create a form of artificial intelligence capable of recognizing objects from fragments, just like our brains do, as well as offering low-power computing, among other interesting features.
said Michael Saccone, of the Los Alamos National Laboratory in the US, who developed the system with colleagues at Aalto University in Finland. “Our paper lays the foundation that we need to use these physical systems in a practical way.”
Winners of the 2016 Nobel Prize in Physics, the rotating glasses sit at the intersection of materials engineering and computing.
It consists of a kind of disordered system of nanomagnets that randomly interact with each other, generating two types of magnetic arrangement in the material. These magnets exhibit a “frustration” phenomenon, meaning that they do not settle into a uniformly structured configuration when the temperature drops, and they have distinct dynamic and electrostatic properties that can be exploited for computing applications.
Now, for the first time, researchers have found a way to make it continuously.
Spinning glasses are a way of thinking mathematically about the structure of matter. What the researchers did was use electron beam lithography to adjust the interaction of atoms within these systems.
This makes it possible to represent a variety of computational problems in physical atomic latticework of spinning glass. That is, you do not need to create hardware and create software that works on it – the hardware already embodies the algorithm itself.
“Theoretical models that describe spinning glasses are widely used in other complex systems, such as those that describe brain function, error-correcting codes, or stock market dynamics,” Saccone said. “This strong interest in spin glass provides a strong impetus for the generation of industrial spin glass.”
The team then combined theoretical elements with experimental tests to make and monitor the artificial glass. The chosen prototype reproduces a Hopfield neural network, which mathematically models the associative memory to direct the perturbation of the system. Spinning glasses and Hopfield networks are emerging fields that develop symbiotically, one feeding off the other.
Associative memory, whether in a Hopfield network or other forms of neural networks, connects two or more memory patterns related to an object. If only one memory is activated, for example, when a partial image of a face is received as input, the network can retrieve the entire face. Unlike more traditional algorithms, associative memory does not require an exactly identical scenario for memory selection.
Another advantage is that the memories of these networks are less disturbed by noise than other neural networks.
According to the team, Artificial Intelligence algorithms developed on rotating glasses will be “more complicated” than traditional algorithms, but they will also be more flexible for some applications.