Artificial neurons are increasingly reminiscent of the human brain compared to traditional computers. This similarity is becoming even clearer as artificial neurons now require sleep to function more efficiently.
Moreover, as a recent study shows, they require not a simple shutdown, but the impact of special slow signals, such as those observed in a sleeping brain.
Neural networks are made up of artificial neurons that exchange signals with each other, like real biological neurons. Usually, the formed connections are strengthened over time, so that neural networks can learn on their own.
Unlike sequential processing of information by conventional computers, neural networks can process different streams of information in parallel, making them a powerful tool for tasks such as image and speech recognition. Curiously, AI "inherited" from a living brain the need for sleep, which is so necessary for our health and well-being.
In a new study conducted by specialists at the Los Alamos National Laboratory (USA), it was found that neurons become less stable after prolonged work. This happened during the uncontrolled mastering of the dictionary by the AI system. Artificial intelligence "overextended" the fact that in the process of identifying similarities between objects for the sake of their classification, it was not provided with clear examples for comparison.
To help the networks maintain their activity, the researchers exposed them to various types of white noise signals (noise, the spectral components of which are evenly distributed over the entire range of frequencies involved - ed. Techcult).
The so-called Gaussian noise - signals that include a wide range of frequencies and amplitudes - is best suited to "calm down" neurons. Most interestingly, these are the same types of waves that propagate through the human brain during the recovery phase of slow wave sleep.