Science & Technology, Australia (Commonwealth Union) – Resembling a set of self-arranged ‘Pick Up Sticks,’ the network of nanowires emulates the synaptic function of the human brain according to researchers. In this groundbreaking experiment, the network was trained to access real-time, dynamic online data and then learn and retain this information.
For the very first time, a physical neural network has successfully demonstrated the capability to learn and remember information on the fly, mirroring the functioning of neurons in the human brain.
This achievement paves the way for the development of energy-efficient machine intelligence suitable for handling more complex, real-world learning and memory tasks.
Published in Nature Communications today, this research is the result of collaboration between scientists from the University of Sydney and the University of California at Los Angeles.
Ruomin Zhu, the lead author and a PhD student at the University of Sydney Nano Institute and School of Physics, indicated that the findings illustrate the way nanowire networks are able to harness brain-inspired learning and memory functions to process dynamic, streaming data.
Researchers of the study pointed out that the nanowire networks consist of minuscule wires with diameters in the billionths of meters. These wires self-organize into patterns reminiscent of the children’s game ‘Pick Up Sticks,’ closely resembling neural networks found in our brains. These networks can be utilized to perform specific information processing tasks.
To achieve memory and learning tasks, simple algorithms respond to changes in electronic resistance at the junctions where the nanowires intersect. This phenomenon, known as ‘resistive memory switching,’ occurs when electrical inputs encounter alterations in conductivity, resembling the behavior of synapses in our brains.
In this study, the network was employed to recognize and remember sequences of electrical pulses that correspond to images, drawing inspiration from the human brain’s information processing methods.
The supervising researcher, Professor Zdenka Kuncic, likened the memory task to recalling a phone number. Additionally, the network was utilized to successfully perform a benchmark image recognition task by accessing images from the MNIST database of handwritten digits, which is a collection of 70,000 small grayscale images commonly used in machine learning.
She further pointed out that their earlier research confirmed the capacity of nanowire networks to retain straightforward tasks. This study has expanded upon those discoveries by demonstrating that tasks can be executed with dynamic data retrieved from online.
“This is a significant step forward as achieving an online learning capability is challenging when dealing with large amounts of data that can be 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.
“Our novel approach allows the nanowire neural network to learn and remember ‘on the fly’, sample by sample, extracting data online, thus avoiding heavy memory and energy usage.”
Mr. Zhu pointed out additional benefits of processing information online, noting, that in cases where data is continuously streamed, as is often the case with sensors, machine learning relying on artificial neural networks must adapt in real-time, a capability they are currently not finely tuned for.
The study showcased the nanowire neural network’s impressive machine learning prowess, achieving an accuracy rate of 93.4% in correctly identifying test images. Moreover, the memory task encompassed recalling sequences of up to eight digits. In both tasks, data was continually streamed into the network to illustrate its aptitude for online learning and to exemplify the synergistic relationship between memory and the learning process.
Machine learning is already transforming industries and the way we live and work. Its evolution promises to bring even more innovations and opportunities in the future. As the field continues to mature, which was evident in this study collaboration between researchers will be crucial.
