Many times we say that the human being is not a machine, needing recreation and sleep to be able to recover from working days. But even, in the same machines, rest is essential for greater efficiency. It is the case of the Artificial intelligence: requires sleep to learn new tasks.
In humans, sleep helps to consolidate what has been learned. The AI, for its part, can learn and remember how to multitask with its kind of artificial sleep.
Maxim Bazhenov, an expert at the University of California San Diego, spoke with New Scientist about.
“Now there is a big trend to bring ideas from neuroscience and biology to improve existing machine learning, and sleep is one of them,” says Bazhenov.
Artificial Intelligence is the attempt to transfer human intelligence to computers, machines or applications. This requires machine and deep learning, offering solutions to problems.
Learning is achieved through training. For example, if it is an AI that produces images, it needs to be fed hundreds of images, to take ideas from them and create new ones. Depending on the level, it goes from automatic to deep: in the latter, it no longer requires human intervention.
How sleep favors Artificial Intelligence
Bazhenov and his colleagues sought to demonstrate how sleep aids Artificial Intelligence. They trained a neural network of spikes, connected to artificial neurons that resemble the structure of the human brain.
The goal was for the AI to learn to do two different tasks without overwriting the connections learned from the first task. And they did it, but interspersing periods of training with periods of sleep.
That is, the AI learned to perform two tasks optically, always giving it a rest time between one and the other.
Hava Siegelmann, team member, explained: “The goal of lifelong learning AI is to have the ability to combine different experiences intelligently and apply this learning to novel situations, just like animals and humans do.”
The researchers’ next step is to demonstrate the widespread use of spike neural networks with more complex tasks.