‘Edge of Chaos’ opens path to artificial intelligence discovery

Scientists at the University of Sydney and Japan’s National Institute of Materials Science (NIMS) have discovered that an artificial network of nanowires can be tuned to respond in a brain-like manner when electrically stimulated.

An international team led by Joel Hochstetter, together with Professor Zdenka Kuncic and Professor Tomonobu Nakayama found that by keeping the network of nanowires in a brain-like state, “At the Edge of Chaos” has worked to an optimal level.

They say this suggests that the fundamental trait of neural intelligence is physical. And their discovery opens an exciting avenue for the development of artificial intelligence.

The study was published today in Nature communication.

“We used wires 10 micrometers long and up to 500 nanometers thick, arranged randomly on a two-dimensional plane,” said lead author Joel Hochstetter, a doctoral candidate at the University of Sydney’s Nano Institute and the School of Physics.

“When the wires overlap They create electrochemical junctions, such as synapses between neurons,” he said. “We found that electrical signals transmitted through this network automatically find the best path for transmitting information. And this architecture allows the network to ‘remember’ the previous path through the system.”

on the edge of chaos

The research team tested a randomized nanowire network using simulations. to see how it works best for a simple solution.

If the signal that stimulates the network is too low Shows that the route is too predictable and organized. and doesn’t produce a complex enough output to be useful. If the electrical signal overflows the network The output is disorganized and useless in debugging.

The optimal signal for generating useful output is at the edge of this chaotic state.

Professor Kuncic from the University of Sydney said: “Some theories in neuroscience suggest that the human mind can operate under this chaos. “Some neuroscientists think it’s in a state where we achieve peak brain performance.”

Professor Kuncic is Mr Hochstetter’s Ph.D. advisor and is currently a Fulbright scholar at the University of California in Los Angeles, working at the intersection between nanoscience and artificial intelligence.

She said: “What’s really exciting about this result is that it shows that this type of nanowire network can be adapted to a regime with a wide range of brain-like dynamics that can be exploited to optimize performance. data processing”

Overcome the duality of computers

in the nanowire network The junction between the wires allows the system to combine memory and functionality into a single system. Unlike standard computers that separate memory (RAM) and functionality (CPU).

“These junctions act like computer transistors. But it has the added feature of remembering that the signal has traveled that route before. For this reason it is called ‘memristors,'” said Hochstetter.

This memory takes a physical form. where the junction at the intersection between the nanowires acts like a switch. The behavior depends on the response to electrical signals in the past. when using signals across these intersections The tiny silver filament stimulates the junction, allowing current to flow through it.

“This creates a memory network within the random system of nanowires,” he said.

Mr Hochstetter and his team created a physical network model to show how they can be trained to solve simple problems.

“For this study We train the network to convert simple waveforms into more complex waveforms,” Hochstetter said.

In their simulations, they adjusted the amplitude and frequency of the electrical signal to see where the best performance occurred.

“We found that if you hit the signal too late The network does the same thing over and over without learning and developing. If we push it too hard and too fast The network becomes unpredictable and unpredictable,” he said.

University of Sydney researchers are working closely with collaborators at The International Center for Materials Nanoarchictectonics at NIMS in Japan and UCLA, where Professor Kuncic is a visiting Fulbright Scholar, the nanowire system was developed at NIMS and UCLA, and Mr Hochstetter developed the analysis in collaboration with co-author and fellow doctoral student Ruomin. Zhu and Alon Loeffler

reduce energy consumption

Professor Kuncic said that the combination of memory and operations offers significant practical advantages for future artificial intelligence developments.

“The algorithm needed to train the network to know which access points should correspond to the appropriate ‘load’ or data weight. It consumes a lot of energy,” she said.

“The system we are developing eliminates the need for such an algorithm. We just allow the network to develop its own weighting. This means that we only need to worry about the input and output signals. which is a framework called ‘Reservoir calculation’ network, the weights are self-adjusting. which could release a large amount of energy.”

She said this means that future artificial intelligence systems using such networks will have much less power.