Alice chases a rabbit with a pocket watch and enters a fantasy underground world where animals can talk, food can make people grow or shrink, and flamingos are used to play croquet. In the fairy tale "Alice in Wonderland", this fantasy underground world is a dream of the little girl Alice. Have you ever had such a fantastic dream? What do these strange dreams mean to the brain? Inspired by the field of deep learning, neuroscientists have proposed that strange dreams may be a way for the brain to train itself. Strange Dreams Our brains produce strange dreams, perhaps to counter the monotony of daily life with novelty. There is an adaptive logic in this: if an animal behaves too strictly in accordance with the environment, it will sacrifice the ability to generalize, understand, and learn new things. In the field of artificial intelligence, scientists call the phenomenon of a model highly fitting a given data set "overfitting." For example, when a face recognition algorithm is trained on a set of images for too long, the algorithm may start to recognize images based on trees or other objects in the background, which defeats the purpose of face recognition. Face recognition algorithms are supposed to learn general rules, that is, to recognize facial contours without being affected by expressions or environment, but in the case of overfitting, the algorithm simply memorizes the training set. So, is our brain's effort to concoct strange dreams to help us avoid learning to "overfit" with daily life? Erik Hoel, a neuroscientist at Tufts University, thinks this conjecture is sound. In a recent paper, he elaborated on his views. "Mammals are constantly learning, and there is no switch to turn off the process. So it is natural to think that mammals will also encounter the problem of overlearning, which needs to be solved through cognitive homeostasis," Hoel said. The 'overfitted brain hypothesis' states that when the effects of biological learning gradually deviate in a certain direction, the organism needs to fight against it and return cognition to a more optimal homeostasis." In the field of dreams, Hull's views are very unique. They not only explain why strange dreams occur, but also propose a purpose for their existence. Other explanations of dreaming do not really answer why dreams become strange, or only explain them as byproducts of other cognitive and physiological processes. These views avoid those strange dreams and say that there are actually very few truly strange dreams, that is, we can easily overestimate the strangeness of our dreams. Although we are generally more likely to remember strange dreams, research shows that about 80% of dreams reflect normal activities and may be very boring. Related Hypotheses The "continuity hypothesis" of dreams states that dreams are simply a recreation of one's waking life. Indeed, most of our dreams (although most of them we probably don't remember) fall into this category. But the continuity hypothesis doesn't explain why certain things occur more frequently in dreams. For example, many people spend a lot of time in front of a computer screen while awake, working, playing games, watching movies, and reading, but do you often dream of yourself sitting in front of a computer? According to the continuity hypothesis, the proportion of some activities in dreams would reflect their proportion in real life, but this is clearly not the case. Another theory is that dreams exist to help people rehearse events that will happen in the real world, and this theory has been supported by many studies. These studies have found that sleep, especially dreams, are very important for learning and memory. Antti Revonsuo, a cognitive neuroscientist at the University of Skövde in Sweden, has proposed two theories based on this feature of dreams. One of them is the "threat simulation theory," which holds that dreams can provide practice for real-world dangers, which could explain why 70% of dreams are painful. Later, Ravensu expanded this theory to suggest that dreams can provide practice for most situations in real life. These theories can explain why we believe that what we see in dreams is real: because if we don't treat dreams as real, we won't take them seriously, and the effect of learning from them will be weakened. In addition, from the perspective of brain structure, the reason why we regard dreams as reality is that the neural activity of the dorsolateral prefrontal cortex (DLPFC) decreases during dreaming. This brain area can monitor abnormal situations in life. The DLPFC is more active in lucid dreaming (characterized by the dreamer being aware that he is dreaming), which can also confirm this theory. Another theory is that strange dreams are a side effect of brain activity. The "random activation theory" proposes that dreams occur because the forebrain tries to make sense of the jumbled and meaningless messages sent by the back of the brain during sleep. In this view, strange dreams have no function, but the random activity of the brainstem is meaningful. Barbara Jones, a neuroscientist at McGill University, noticed that the brainstem regulates things like sex and running, and similar scenes often appear in dreams. Unlike other hypotheses, Hull's hypothesis confronts the weirdness of dreams and gives them meaning. Hull believes that strange dreams help prevent the brain from overfitting, a problem that also plagues researchers in the field of machine learning. Overfitting refers to paying attention to irrelevant details in the training set, and stopping learning is one way to deal with this situation. A broader approach is to introduce noise, that is, input distorted information. Introducing noise will make it impossible for deep learning neural networks to determine the importance of specific information in the training set, making them more likely to pay attention to general information, so they can work better in the real world. Hull believes that strange dreams have a similar function to the introduction of noise when training neural networks: providing some distorted input to prevent the brain from overfitting waking life and the "training set." What the human brain and algorithms have in common Interestingly, some experiments have confirmed that overfitting occurs in humans, and that sleep can eliminate it. In short, dreams are weird because we need them to be. If dreams are too similar to real life, overfitting will be exacerbated rather than eliminated. In general, even dreams that feel very real are usually not exactly the same as real events. Similar to other ideas that dreaming helps people learn to cope with the real world, Hull's hypothesis suggests that sleep is the best time for "offline" learning. Distorted experiences or distorted inputs can be distracting or even dangerous if they occur when we are awake. And the reason why we forget many dreams may be to avoid confusing them with real things. After all, the brain just wants to use it to train the neural network, and does not want to create new memories to confuse with reality. Can machine learning help us guess the "best" strange dreams? "Maybe," Hull said, "but I prefer the other direction, that deep learning should learn more from neuroscience research. We want to feed the program data that is different enough, but not so outrageous that it exceeds its processing ability." All this suggests that dreams should have a certain "optimal" level of weirdness. But "weirdness" is not an easily measured dimension. "It's a lot like literature and art," Hull said. "For example, a good poem should be neither completely incomprehensible nor too plain. It needs to find the right place for word changes and metaphors. 'Just right' can best help large and complex brains solve a series of problems such as overlearning, overmemory and overfitting." Inspired by the structure of the brain, scientists developed neural networks, but with the development of deep learning, artificial intelligence is mostly used to create smarter machines rather than simulate and understand human thinking. But more and more discoveries in the field of deep learning still inspire us to new theories about how the brain works. In order to learn better, neural networks need to learn some strange and meaningless data. Perhaps, we humans also need it. Reviewer of this article: Chen Haixu, Deputy Director and Master Supervisor of the Second Medical Center of PLA General Hospital Source: Global Science (ID: huanqiukexue) By Jim Davies Translation: Zheng Yuhong Original link: https://nautil.us/blog/weird-dreams-train-our-brains-to-be-better-learners |
<<: Big Questions in Aging Research: What’s the Secret to Centenarians’ Longevity?
>>: Year-end review: Major breakthroughs in Chinese science and technology in 2021
Ideal Auto announced its sales results for Februa...
The so-called traffic means acquiring new users. ...
Recently, the entry #One person controls 600 mobi...
What I am sharing today has been verified by hund...
Author: Xue Qingxin, registered dietitian Reviewe...
Advanced model tree course, solving most learning...
“The opposite of love is not hate, but ignoring.”...
Have you all been flooded with Wu Yifan's Fre...
After creating a category, you must dominate the ...
Complete set of skipping rope teaching videos - C...
Where there is a vent, there will be a "bubbl...
Before we know it, another year has passed, and m...
With the development of mobile Internet, finance ...
Although DSP and information flow have been aroun...
Recently, the global automotive industry has been...