Artificial neurons are becoming more and more like the human brain than traditional computers. This similarity becomes even more obvious, as now the artificial neurons to work more effectively needed sleep.

Moreover, as shown by a recent study, they need not a simple disconnection, and the special effects of the slow signals, similar to those seen in sleep brain.

Neural networks consist of artificial neurons that communicate with each other by signals, like a real biological neurons. Usually formed connection grows stronger, thanks to what a neural network can be trained.

In contrast to the sequential information processing in conventional computers, neural networks can process different streams of information in parallel, which makes them a powerful tool for solving tasks such as image recognition and speech. Curiously, AI has “inherited” from the human brain and the need for sleep, so necessary to our health and well-being.

In a new study conducted by experts at Los Alamos national laboratory (USA), it was found that the neurons become less stable after prolonged operation. It happened during the uncontrolled development of AI system dictionary. Artificial intelligence is “overextended” from the fact that in the process of identifying similarities between objects for their classification, he had not been provided clear examples for comparison.

To help the networks to maintain their activity, the researchers subjected them to the impact of different types of white noise (noise spectral components which are evenly distributed over the entire range of frequencies involved – approx. ed Techcult).

It is best to “pacify” neurons came the so-called Gaussian noise signals, including a wide range of frequencies and amplitudes. Most interesting is that these are the same types of waves that propagate through the human brain during the recovery phase of slow sleep.
Source — Los Alamos National Laboratory