The biological learning methods of habituation and sensitization help organisms, including humans, adapt to changes in their environment. Each can be exemplified in riding a bike. As you learn to ride, you become habituated to the bike’s wobbles and respond less to its tilts and vibrations. You’re able to ride faster and farther as a result. But if you fall and hurt yourself, you’ll be more sensitized to the bike’s instability, and as a result more dependent on the brakes or training tools.
Scientists in the field of neuromorphic computing, which aims to make computers smarter and more independent from human input, are working to reproduce aspects of habituation and sensitization in hardware. A new paper in Proceedings of the National Academy of Sciences demonstrates similar behaviors in a semiconductor called nickel oxide, which could help researchers develop more agile and adaptable devices.
In living organisms, habituation and sensitization can help resolve the stability-plasticity dilemma, whether we decide to hold onto old information (be more stable), or be more responsive to new information (be more plastic). Just as neither the overly cautious cyclist who never gains speed nor the reckless cyclist who never brakes will master the art of riding, machines won’t learn effectively if they can’t determine how to prioritize items within massive amounts of data or information.
But for a computer to learn how to learn better, it must be fundamentally different from modern machines. “The traditional materials used for electronics have not been designed for brain-inspired computing,” says Shriram Ramanathan, Ph.D., a professor at Purdue University and one of the study’s authors.
Ramanathan and his colleagues set out to demonstrate habituation and sensitization in hardware using a quantum substance (anything with properties that can’t be explained by classical physics) called nickel oxide. They repeatedly exposed the semiconductor to various gases and light, and watched how quickly it responded to changes as measured by the material’s electrical resistance.
The team placed nickel oxide into a box at 200 degrees Celsius, then added a gas containing hydrogen. The hydrogen bonded with the nickel oxide’s oxygen atoms. But when the hydrogen gas was turned off, the excess hydrogen released from the nickel oxide as it reacted with oxygen in the air. “You can think of the material as breathing oxygen in and out,” explains Ramanathan. When oxygen left the nickel oxide for the surrounding air, that chemical reaction left electrons behind in the semiconductor. And when the influx of oxygen gas entered the material, it bonded with those extra electrons. Those changes in the availability of electrons, he says, changed how easy it was for an electrical current to travel within the nickel oxide.
What We’re Learning From Sea Slugs’ Giant Neurons
Shriram Ramanathan and his colleagues modeled their nickel oxide experiments after experiments previously conducted with sea slugs in the genus Aplysia. For centuries, scientists have used these slugs to study learning, behavior,
The slugs can retain information for weeks—a killer “long-term” memory for an organism with a year-long life span—and display a relatively high level of neural plasticity. It may not make them “intelligent,” but it does make them the perfect subject for researchers teasing out the origins of fundamental learning behaviors like habituation and sensitization. Aplysia’s colossal neurons, the largest in the animal kingdom, can grow to 1 mm in length. The neurons’ gargantuan size make it easier for scientists to physically see and study physiological responses to stimuli.
The team repeatedly added and withdrew the hydrogen gas, reducing the time between exposures, to see how this reaction changed over time. The first time the nickel oxide was exposed to hydrogen gas, it reacted strongly. After 15 minutes, around 99.95 percent less electrical current was passing through. But if scientists shortened the period between hydrogen exposures to between 45 and 15 seconds, they found there wasn’t as much oxygen available to react with the second round of hydrogen. That meant the electrical resistance didn’t change as much during subsequent rounds—mirroring the reaction-damping effect of habituation.
Researchers then sensitized the material using ozone, a highly reactive gas made of three oxygen molecules. When immersed in ozone, the nickel oxide took in oxygen quickly, decreasing its electrical resistance. Next time it was exposed to hydrogen gas, there was more oxygen to attract the hydrogen—and a stronger change in electrical resistance.
“This is an exact analog to biological systems,” says Hai (Helen) Li, Ph.D., a professor of electrical and computer engineering at Duke University, who was a Ph.D. student of one of the co-authors but did not work on the study. “[Humans] have sensors and receive the external signals, and then we process it immediately; it doesn’t have to pass through something else.” Right now, machines use different systems to receive information, then process it—like how a camera takes in light and then processes it into a photo. But in this study, the changes in the nickel oxide’s electrical resistance were a direct reaction to the gases in its environment. Neuromorphic computing could similarly enhance a computer’s ability to receive and process information—when applied to devices like cameras, these machines could become smaller and more efficient. More generally, says Li, our brains use very little power in comparison to computers; successfully mimicking the brain’s processes could therefore save energy.
Other devices could use the technology to self-tailor to the consumers’ needs. Medical implants with neuromorphic computing, says Li, could send signals along an injured nerve to help people regain use of paralyzed fingers or toes. Independent from human programmers, such a device would make its own decisions based on biological fluctuations.
Consumer products will require greater advancement in the neuromorphic computing field than this sole study, however. Here, habituation and sensitization appeared as chemical reactions to hydrogen and ozone, but computers run on an electric current rather than hydrogen or ozone gas. “Being able to do all of this by electrical stimulus will be fascinating,” Ramanathan says. “Then you can start to use traditional stimulus that is historically used in electronics.”
This content is created and maintained by a third party, and imported onto this page to help users provide their email addresses. You may be able to find more information about this and similar content at piano.io