Researchers at Los Alamos National Laboratory are using a rare form of matter known as spin glass to replace circuits. The unusual properties of spin glass enable a form of AI that can recognize objects from partial images as the brain does. “Spin glasses are systems with a ‘bumpy landscape’ of possible solutions,” Cris Moore, a computer scientist and physicist at the Santa Fe Institute, who was not involved in the Los Alamos research, told Lifewire in an email interview. “They help us analyze why algorithms sometimes get stuck in solutions that look good locally but are not the best possible.”
Printable Circuits
The use of spin glass for printable circuits could also lead to new types of low-power computing. The spin-glass allows researchers to investigate material structures using mathematics. With this approach, scientists can tweak the interaction within systems using electron-beam lithography, which uses a focused beam of electrons to draw custom shapes on a surface. The lithography could allow the printing of new types of circuitry. The lithography makes it possible to represent a variety of computing problems in spin-glass networks, according to a recent paper by the Los Alamos team published in the peer-reviewed journal Nature Physics. “Our work accomplished the first experimental realization of an artificial spin-glass consisting of nanomagnets arranged to replicate a neural network,” Michael Saccone, a post-doctoral researcher in theoretical physics at Los Alamos National Laboratory and lead author of the paper, said in the news release. “Our paper lays the groundwork we need to use these physical systems practically.” Moore likened spin glass to silicon dioxide (window glass), which appears to be a perfect crystal, but as it cools, it gets stuck in an amorphous state that looks like a liquid on a molecular level. “In the same way, algorithms can get stuck behind ’energy barriers’ that stand in the way of the global optimum,” Moore added. Ideas from spin glass theory could help researchers navigate high-dimensional landscapes. “This pursuit has created a vibrant interdisciplinary community at the intersection of physics, mathematics, and computer science,” Moore said. “We can use ideas from physics to determine fundamental limits on algorithms—like how much noise they can tolerate while still finding patterns in data—and to design algorithms that succeed all the way up to those theoretical limits.”
AI That Remembers Like Humans
The research team investigated artificial spin glass as a way to look into what are called Hopfield neural networks. These networks model human associative memory, which is the ability to learn and remember the relationship between unrelated items. With associative memory, if just one memory is triggered, for example by receiving a partial image of a face as input—then the network can recall the entire face. Unlike traditional algorithms, associative memory doesn’t require an identical scenario to identify a memory. The research by Saccone and the team confirmed that spin-glass will be helpful to describe the properties of a system and how it processes information. AI algorithms developed in spin glass would be “messier” than traditional algorithms, Saccone said, but also more flexible for some AI applications. “Theoretical models describing spin glasses are broadly used in other complex systems, such as those describing brain function, error-correcting codes, or stock-market dynamics,” Saccone said. “This wide interest in spin glasses provides strong motivation to generate an artificial spin glass.” Other types of brain-inspired chips could also improve how AI recognizes images. A recent paper shows how computer chips could dynamically rewire themselves to take in new data like the brain does, helping AI to keep learning over time. “The brains of living beings can continuously learn throughout their lifespan,” Shriram Ramanathan, a professor in Purdue University’s School of Materials Engineering and one of the paper’s authors said in a news release. “We have now created an artificial platform for machines to learn throughout their lifespan.”
