Physicists have created an electrical network that can learn

A team of scientists, led by physicist Samuel Dillavue, assembled a small electrical grid, connecting

The researchers set the voltage for specific input nodes and read theBy adjusting the resistors on its own, the network learned to produce the desired data for a given set of input values.

“The network has been configured to perform manysimple AI tasks,” says Dillavu. “For example, it can distinguish between three types of iris flowers with more than 95% accuracy based on four parameters: the length and width of the petals and sepals.”

For machine learning, AI is usuallyusing artificial neural networks. Such networks usually exist only in the computer's memory. A neural network consists of points or nodes, each of which can take on a value from 0 to 1, connected by edges. Each edge has its own weight depending on the values ​​in the nodes. When training such a system, it is necessary to adjust the weight of the edges in order to obtain the desired result.

“This is a tricky optimization problem thatincreases significantly with the size of the network and requires a large amount of computing resources,” notes Dillavu. “The situation is complicated by the fact that all the edges must be tuned at the same time.”

To get around this problem, physicists looked for systems that could tune themselves without external calculations.

In their research, the scientists built two identical networks on top of each other.In a closed network, they applied voltage and fixed the desired values on the output elements.In an open network, only the voltage across the input resistor was set.

The system regulated the resistance on the resistorsin two networks depending on the voltage difference between identical nodes in each of them. Over several iterations, these adjustments brought all voltages across all resistors in the two networks into line. The system has learned to produce the correct output for given input values.

Photo: Science

"This setup requires little computation,Dillavu says. — The system only needs to compare the voltage drop across the respective resistors in the closed and free networks using a comparator. Our work proves the fundamental possibility of a new way of machine learning that does not require large calculations.”

Read more:

"James Webb" took the clearest photo of a star in history

Voluntary death. We tell how the euthanasia procedure works around the world

Scientists stumble upon 1,500-year-old archaeological anomaly