Neural networks helped answer the question of why we dream

Our world is full of superstitions, some of themare of particular interest. It is known, for example, that Michael Jordan wore University of North Carolina shorts under his Chicago Bulls uniform; Serena Williams wears the same socks throughout the tournament; and tennis player Goran Ivanishevich at one time watched the cartoon "Teletubbies" throughout the competition (later even he could not answer why). For some, this is nothing more than just "cockroaches" in the head, but scientists believe that they appear for a reason. The new theory not only explains what this behavior is associated with, but also allows you to answer the question, why do you even dream.

Not everyone has the ability to dream, but it is very important.

Content

• 1 Where do superstitions come from
• 2 What are dreams?
• 3 Why do we dream
• 4 What is the danger of lack of sleep?

Where do superstitions come from?

Psychologists say that this behavior occursdue to the fact that the human brain sometimes connects events that in fact have little or no causality. Machine learning researchers have a different answer to this question. This is an example for them "Retraining" - using irrelevant parts forbuilding a neural network. There can be many factors that contribute to the success of a particular tennis serve or basketball throw, but the color of the socks or underpants is hardly one of them.

The exact same thing happens with artificialneural networks. During the training, they learn important details, but also irrelevant ones that are simply not needed. Overfitting is a bane for machine learning experts who have developed a wide variety of techniques to get around it.

If the neural network is fed too much unnecessary data, it will produce something like this. Some have similar dreams

But if the neural network copes with retraining, then how are things going? in the human brain?

This question can be answered thanks to new work by Eric Hoel, a neuroscientist at Tufts University in Massachusetts. According to his theory, the human brain prevents overfitting. with the help of dreams... Dreams were developed specifically to addressthis problem, which is common to all neural networks. If his theory is correct, it provides an answer to one of neuroscience's big unsolved problems: why do we even dream.

What are dreams?

A bit of background. Psychologists, neuroscientists, and other scientists have pondered the origin and role of dreams since time immemorial. Freud suggested that they were a way of expressing frustration when a person did not implement one of his ideas.

Others believe that dreams are a kind ofan emotional tool that allows us to control and resolve emotional conflicts. However, critics point out that most dreams lack strong emotional content, and meaningless dreams without any emotion are common.

Still others say that dreams are part of the process,which the brain uses to rectify memories or selectively forget unwanted or unnecessary memories. However, this is also not entirely true, since most dreams lack realistic detail, they have strange orientations, similar to hallucinations, and often contain previously unknown information. Have you had a meaningful dream for a long time? Tell us in our Telegram chat.

Most dreams have nothing to do with specific memories at all, making integrating new memories a dubious dream goal, says Hoel.

Why do we dream

His new idea is that the purpose of dreams is help the brain make generalizationsbased on specific experience. Build real connections. And dreams do it the same way machine learning specialists prevent overfitting in artificial neural networks.

Dreams help the brain to get rid of unnecessary connections and "clear its head"

How do they do it? The most common way is to add a little "noise" to the training process so that it is harder for the neural network to focus on irrelevant details. In practice, researchers add "noise" to images or download corrupted data into a computer, and even remove random nodes in the neural network. From a human point of view, this would be tantamount to making Michael Jordan wear different shorts or making Serena Williams change socks. This would greatly reduce the likelihood of them focusing on a particular, irrelevant detail — in their case, superstition.

Dreams have the same function for the brain, Hoel said. He calls this idea the converted brain hypothesis and indicates that there are manyevidence. For example, one of the best ways to induce dreams is to play simple, repetitive games like Tetris. This creates an environment in which the brain can become “retrained” with many new repetitive details.

This is why such actions induce dreams. These dreams are not repetitions of Tetris games, and tend to contain few details more like hallucinations. It is this “noise” that helps the brain to remove unnecessary connections. This is why people can improve their vital signs after a good night's sleep.

The scientist uses his new theory to make a number of bold assumptions.

Perhaps the human brain can measure overfitting, he says.

Dreams help you think clearly

What is the danger of lack of sleep?

The theory can also be used for betterunderstanding the types of mistakes people with sleep deprivation can make. If the "sleep deprived" brain is overwhelmed with unnecessary connections, it will tend to make stereotypical mistakes. So Michael Jordan and Serena Williams might just have had a good night's sleep.

Also, why people create and enjoy fiction has always been a mystery. But Hoel has an answer:

The refitted brain hypothesis suggests that fiction, and perhaps art in general, may have basic cognitive utility in preventing overfitting, since they act like artificial dreams.

Interesting job! Until now, most cognitive theories have viewed dreams as an epiphenomenon, a by-product of general sleep that has no meaningful function of its own. This work turns it all upside down, revealing the biological function of dreams and, therefore, for the first time substantiating their evolution.