Advancing AI Applications Through Neuromorphic Computing Research

6 months ago by Luke James

The University of Sydney, in collaboration with UCLA and the Japanese National Institute of Materials Science, has developed a neuromorphic network that exhibits emergent brain-like behavior by resembling some cognitive functions.

The brain is believed by many to be the most complex structure in the universe; it is by far the most mysterious and least understood. To give you an idea of just how little we understand about the human brain, far more is known about our own galaxy – something that stretches for tens of thousands of lightyears or more – than is known about the chunk of grey matter nestled safely within our skulls.

And although intelligence – the ability to acquire, retain, and apply knowledge and skills – may be one of the human brain’s most fascinating properties, relatively little is known about how this works: We do not know how we remember, learn, and think. 

This is a major reason behind science’s – and indeed the wider world’s – fascination by artificial intelligence (AI). AI has shown us that it is possible to emulate certain parts of what we know or believe to be ‘intelligence’ through the feeding of data into a computer and the use of algorithms to find, remember and recognize patterns and information that can be used to make predictions that grow in accuracy over time – machine learning.

Now, new AI systems are being developed by researchers at the University of Sydney, in a bid to bring improve the power of AI and potentially develop systems that stretch beyond its reach.

 

An optical micrograph of a nanowire network developed by Sydney University researchers.

An optical micrograph image of the nanowire network created by the partnership between U.S., Australian, and Japanese researchers. Image Credit: The University of Sydney. 

 

A Synthetic Neural Network Using Nanotechnology

The University of Sydney Nano Institute and School of Physics is currently working alongside international researchers to develop a synthetic neural network using nanotechnology, with the potential to develop new systems in AI and machine learning. 

Led and funded by the International Centre for Materials Nanoelectronics at the National Institute of Materials Science in Japan, the project is also being contributed to by the California NanoSystems Institute at UCLA

 

Development Process of the Neural Network 

The researchers’ neuromorphic network – a large-scale artificial system that mimics biological functions of the nervous system – was created using self-assembled nanowires that form contacts between adjacent nanowires. Each contact between the nanowires exhibits a synaptic-like response similar to those seen at the junction of nerve cells.

The nanowires were made of a silver and polymer composite material, with their average width measuring in at 360 nanometers. Using this network, the research team investigated electrical signal transmission across preferred paths in the network. When electrically stimulated, the team found that the neuromorphic network exhibited emergent brain-like behavior resembling cognitive functions such as learning, memorization, and forgetting. 

Professor Zdenka Kuncic from Sydney Nano and the School of Physics said, “This is exciting because it opens up the possibility of processing dynamically changing data that existing machine learning and AI methods can’t handle.”

 

Developing AI Using Synthetic Neural Networks

The team’s investigations into electrical signal transmission revealed continuous fluctuations that allowed them to exploit multiple transport pathways across the network and spontaneously adapt to changing transmission routes. 

Based on this, the team is now said to be developing next-generation memory devices and neuromorphic information processing systems using these nanowire networks. This has the potential to open up new possibilities for neuromorphic information processing technologies that will benefit AI systems and hopefully lead to new data processing capabilities beyond the reach of conventional AI. 

Comments