The question of why we dream is a subject of division within the scientific community: it is difficult to prove concretely why dreams happen and the field of neuroscience is saturated with hypotheses. Inspired by the techniques used to train deep neural networks, Erik Hoel, research assistant professor of neuroscience at Tufts University, argues for a new theory of dreams: the over-adjusted brain hypothesis. The hypothesis, described on May 14 in a review of the journal Reasonssuggests that the strangeness of our dreams serves to help our brains better generalize our daily experiences.
“There are obviously an incredible number of theories as to why we dream,” says Hoel. “But I wanted to draw attention to a dream theory that takes the dream itself very seriously – that says the experience of dreams is the reason you dream.”
A common problem with AI training is that she becomes too familiar with the data she is trained on – she begins to assume that the training set is a perfect representation of all that he could. meet. Data scientists correct this problem by introducing chaos into the data; in one of these regularization methods, called “abandonment”, some data is ignored at random. Imagine if black boxes suddenly appeared on the internal screen of a self-driving car: the car that sees the random black boxes on the screen and focuses on the overall details of its surroundings, rather than the specifics of that driving experience. particular driving, will probably be better. understand the general driving experience.
“The original inspiration for deep neural networks was the brain,” says Hoel. And while comparing the brain to technology isn’t new, he explains that using deep neural networks to describe the over-equipped brain hypothesis was a natural connection. “If you look at the techniques that people use to regularize deep learning, it often happens that these techniques have striking similarities to dreams,” he says.
With that in mind, his new theory suggests that dreams succeed in making our understanding of the world less simplistic and more complete – because our brains, like deep neural networks, are also becoming too familiar with the “training set” of. our daily. Lives. To counter familiarity, he suggests, the brain creates a strange version of the world in dreams, the mind’s version of dropping out. “It is the very strangeness of dreams in their divergence from waking experience that gives them their biological function,” he writes.
Hoel says there is already evidence from neuroscience research to support the over-equipped brain hypothesis. For example, the most reliable way to spark dreams about something that happens in real life has been shown to be repetitively performing a new task while you are awake. He argues that when you over-train on a new task, the condition of overfitting is triggered and your brain then tries to generalize that task by creating dreams.
But he believes there is also research that could be done to determine if this is really what we dream about. He says well-designed behavioral tests could tell the difference between generalization and memorization and the effect of sleep deprivation on the two.
Another area he wishes to explore is that of “artificial dreams”. He came up with an over-adjusted brain hypothesis when thinking about the purpose of works of fiction like movies or novels. Now he hypothesizes that external stimuli like novels or television shows might act as dream “substitutions” – and that they might even be designed to help delay the cognitive effects of deprivation. of sleep with an emphasis on their dreamlike nature (for example, in virtual reality technology).
While you can just turn off learning in artificial neural networks, says Hoel, you can’t do that with a brain. Brains are always learning new things – and that’s where the over-equipped brain hypothesis comes in. “Life is boring sometimes,” he says. “Dreams are there to keep you from becoming too modeled on the world.”
Source of the story:
Material provided by Cell press. Note: Content can be changed for style and length.