Tiny networks that arrange themselves like the classic game Pick-up Sticks are behind a critical advance for machine intelligence, scientists say.
Inspired by the way the human brain processes information, researchers from the University of Sydney Nano Institute have used a network of wires to recognise and remember sequences of electrical pulses that correspond to images.
Previously, the technology has not been ready for remembering and learning in real-time.
Physicist Ruomin Zhu said if data was being streamed continuously, for example from a sensor, machine learning that relied on artificial neural networks would need to have the ability to adapt.
"The findings demonstrate how brain-inspired learning and memory functions using nanowire networks can be harnessed to process dynamic, streaming data," Mr Zhu said.
The networks are made up of tiny wires that are mere billionths of a metre in diameter, which mimic the brain's neural networks and can be used to process information.
Memory and learning tasks are achieved using a set of commands, or algorithms, that respond to changes in electronic resistance where the nanowires overlap, like the junctions of sticks in the children's game.
Supervising researcher Professor Zdenka Kuncic said the latest memory task was similar to remembering a phone number and involved recalling sequences of up to eight digits.
Their previous research established the ability of nanowire networks to remember simple tasks.
Prof Kuncic said the latest research with the University of California at Los Angeles (UCLA) has extended these findings by showing a machine can deal with large amounts of data that are continuously changing.
"A standard approach would be to store data in memory and then train a machine learning model using that stored information. But this would chew up too much energy for widespread application," she said.
"Our novel approach allows the nanowire neural network to learn and remember 'on the fly', sample by sample, extracting data online."
The researchers said this avoids heavy energy and memory usage, opening the door for efficient and low-energy machine intelligence for more complex, real-world learning and memory tasks by machines.
Their research paper was published in Nature Communications.
