Using electricity to find materials that can “learn” Artificial intelligence and robotics news

Scientists used the advanced photon source to observe behavior mimicking non-living material associated with learning, paving the way for better artificial intelligence.

Scientists seeking to create a new generation of supercomputers seek inspiration from the most complex and power-efficient computer ever built: the human brain.

In some of their early forays into building brain-inspired computers, researchers are investigating different non-biological materials whose properties could be matched to show evidence of learning behaviors. These materials could form the basis of hardware that could be combined with new software algorithms to enable more powerful, useful and energy-efficient artificial intelligence (AI).

In a new study by scientists at Purdue University, researchers exposed oxygen-deficient nickel oxide to brief electrical pulses and elicited two different electrical responses similar to learning. The result is an all-electric system that exhibits these learning behaviors, said Rutgers University professor Shriram Ramanathan. (Ramanathan was a professor at Purdue University at the time of this work.) The research team used resources from the Advanced Photon Source (APS), a user facility of the U.S. Department of Energy’s Office of Science ( DOE) at DOE’s Argonne National Laboratory.

The first response, habituation, occurs when the material “gets used to” being lightly zapped. The scientists noticed that although the resistance of the material increases after an initial jolt, it quickly gets used to the electrical stimulus. “Addiction is like what happens when you live near an airport,” said Fanny Rodolakis, physicist and beamline scientist at APS. “The day you move in, you think ‘what a racket’, but in the end you hardly notice it anymore. »

The other response shown by the material, sensitization, occurs when a greater dose of electricity is administered. “With a larger stimulus, the material’s response increases rather than decreases over time,” Rodolakis said. “It’s like watching a horror movie and then having someone say ‘Boo!’ from behind a corner – you really see it jump. »

“Almost all living organisms exhibit these two characteristics,” Ramanathan said. “They are really a fundamental aspect of intelligence. »

Both of these behaviors are controlled by quantum interactions between electrons that cannot be described by classical physics and which help to form the basis of a phase transition in the material. “An example of a phase transition is a liquid becoming a solid,” Rodolakis said. “The hardware we’re looking at is right on the edge, and the competing interactions taking place at the electronic level can easily be swung one way or another by small stimuli. »

Having a system that can be completely controlled by electrical signals is essential for brain-inspired computing applications, Ramanathan said. “Being able to manipulate materials in this way will allow the material to take on some of the intelligence responsibility,” he explained. “Using quantum properties to embed intelligence into hardware represents a key step towards energy-efficient computing. »

The difference between habituation and sensitization may help scientists overcome a challenge in AI development called the stability-plasticity dilemma. Artificial intelligence algorithms can often be, on the one hand, too reluctant to adapt to new information. But on the other hand, when they do, they can often forget some of what they have already learned. By creating a material that can be habituated, scientists can teach it to ignore or forget unnecessary information and thus gain additional stability, while awareness could train it to remember and incorporate new information, thus allowing plasticity.

“AI often struggles to learn and store new information without overwriting already stored information,” Rodolakis said. “Too much stability prevents AI from learning, but too much plasticity can lead to catastrophic forgetting. »

One of the main advantages of the new study was the small size of the nickel oxide device. “This kind of learning hadn’t been done before in today’s generation of electronics without large numbers of transistors,” Rodolakis said. “This single-junction system is the smallest system to date to show these properties, which has great implications for the possible development of neuromorphic circuits. »

To detect the atomic-scale dynamics responsible for habituation and sensitization behaviors, Hua Zhou of Rodolakis and Argonne used X-ray absorption spectroscopy at beamlines 29-ID-D and 33-ID-D of the APS.

An article based on the study was published in the Steptember 19 issue of Advanced intelligent systems.

The research was funded by the DOE’s Office of Science (Office of Basic Energy Sciences), the Army Research Office, the Air Force Office of Scientific Research and the National Science Foundation.

We want to thank the writer of this write-up for this remarkable material

Using electricity to find materials that can “learn” Artificial intelligence and robotics news


You can find our social media pages here and additional related pages here.https://www.ai-magazine.com/related-pages/